hotfix: Extend Tesseract instead of Tesseract1

Merge branch 'RED-8155' into 'master'
RED-8155: bold-detection in ocr-service Closes RED-8155 See merge request redactmanager/ocr-service!33
2024-01-15 16:13:43 +01:00 · 2024-01-08 13:53:30 +01:00 · 2024-01-08 13:49:58 +01:00 · 2024-01-05 16:05:53 +01:00 · 2024-01-05 16:05:53 +01:00 · 2023-12-22 15:14:42 +01:00
47 changed files with 1611 additions and 476 deletions
@@ -10,7 +10,7 @@ deploy:
  script:
    - echo "Building with gradle version ${BUILDVERSION}"
    - gradle -Pversion=${BUILDVERSION} publish
-    - gradle bootBuildImage --cleanCache --publishImage -PbuildbootDockerHostNetwork=true -Pversion=${BUILDVERSION}
+    - gradle bootBuildImage --publishImage -PbuildbootDockerHostNetwork=true -Pversion=${BUILDVERSION}
    - echo "BUILDVERSION=$BUILDVERSION" >> version.env
  artifacts:
    reports:
@@ -15,11 +15,17 @@ The service uses PDFTron to attempt the removal of invisible elements and waterm
  Extracts all images from the PDF using PDFBox
 3. Striped Image Detection and Stitching  
  Detects if images are striped and stitches them together using Ghostscript.
-4. Binarization  
-  Binarizes the resulting images using Leptonica and the Otsu thresholding algorithm.
+4. Image Processing
+   - Convert to grayscale
+   - Upscale to target DPI
+   - Filter using Gauss kernel
+   - Binarizes the resulting images using Leptonica and the Otsu thresholding algorithm.
+   - Despeckle using various morphological operations 
 5. OCR Processing  
 Runs Tesseract on the images to extract text.
-6. Text Integration  
+6. Font style detection  
+Detection of bold text using stroke width estimation
+7. Text Integration  
 Draws the resulting text onto the original PDF using PDFBox.

 Steps 2.-5. happen in parallel and communicate via a blocking queue to limit RAM usage. 
@@ -25,6 +25,8 @@ tasks.named<Test>("test") {
    reports {
        junitXml.outputLocation.set(layout.buildDirectory.dir("reports/junit"))
    }
+    minHeapSize = "512m"
+    maxHeapSize = "8192m"
 }

 tasks.test {
@@ -14,13 +14,14 @@ dependencies {
    api("net.sourceforge.tess4j:tess4j:5.8.0")
    api("com.iqser.red.commons:metric-commons:2.1.0")
    api("com.iqser.red.commons:storage-commons:2.45.0")
-    api("com.knecon.fforesight:tenant-commons:0.13.0")
+    api("com.knecon.fforesight:tenant-commons:0.19.0")
    api("com.pdftron:PDFNet:10.5.0")
    api("org.apache.pdfbox:pdfbox:3.0.0")
    api("org.apache.pdfbox:jbig2-imageio:3.0.4")
    api("com.github.jai-imageio:jai-imageio-core:1.4.0")
    api("com.github.jai-imageio:jai-imageio-jpeg2000:1.4.0")
-    api("io.github.karols:hocr4j:0.1.2")
+    api("org.apache.commons:commons-math3:3.6.1")
+    api("io.github.karols:hocr4j:0.2.0")
    api("com.amazonaws:aws-java-sdk-kms:1.12.440")
    api("com.google.guava:guava:31.1-jre")
    api("com.iqser.red.commons:pdftron-logic-commons:2.20.0")
@@ -0,0 +1,36 @@
+package com.knecon.fforesight.service.ocr.processor.model;
+
+import java.awt.geom.Rectangle2D;
+import java.awt.image.BufferedImage;
+
+import org.apache.pdfbox.pdmodel.graphics.color.PDColorSpace;
+import org.apache.pdfbox.util.Matrix;
+
+import com.knecon.fforesight.service.ocr.processor.utils.ImageProcessingUtils;
+
+import lombok.AccessLevel;
+import lombok.Getter;
+import lombok.RequiredArgsConstructor;
+import lombok.SneakyThrows;
+import lombok.experimental.FieldDefaults;
+import net.sourceforge.lept4j.Pix;
+import net.sourceforge.lept4j.util.LeptUtils;
+
+public record ExtractedImage(
+        int pageNumber, QuadPoint position, int height, int width, BufferedImage image, Matrix ctm, int numberOnPage, PDColorSpace colorSpace) implements UnprocessedImage {
+
+    @SneakyThrows
+    public Pix asPix() {
+
+        BufferedImage image = ImageProcessingUtils.convertToDeviceColorSpace(this);
+        ImageProcessingUtils.setAlphaChannelToWhite(image);
+        return LeptUtils.convertImageToPix(image);
+    }
+
+
+    public QuadPoint getImageCoordinatesInInitialUserSpace() {
+
+        return QuadPoint.fromRectangle2D(new Rectangle2D.Double(0, 0, 1, 1)).getTransformed(ctm.createAffineTransform());
+    }
+
+}
@@ -1,16 +1,16 @@
 package com.knecon.fforesight.service.ocr.processor.model;

-import java.awt.AlphaComposite;
-import java.awt.Color;
-import java.awt.Graphics2D;
-import java.awt.Transparency;
+import java.awt.Graphics;
 import java.awt.geom.AffineTransform;
 import java.awt.image.BufferedImage;
-import java.io.IOException;
-import java.nio.IntBuffer;

+import org.apache.pdfbox.pdmodel.graphics.color.PDDeviceGray;
+import org.apache.pdfbox.pdmodel.graphics.color.PDDeviceRGB;
 import org.apache.pdfbox.util.Matrix;

+import com.knecon.fforesight.service.ocr.processor.service.threads.OCRThread;
+import com.knecon.fforesight.service.ocr.processor.utils.ImageProcessingUtils;
+
 import lombok.AccessLevel;
 import lombok.Getter;
 import lombok.RequiredArgsConstructor;
@@ -18,107 +18,25 @@ import lombok.Setter;
 import lombok.SneakyThrows;
 import lombok.experimental.FieldDefaults;
 import lombok.extern.slf4j.Slf4j;
-import net.sourceforge.lept4j.Leptonica1;
 import net.sourceforge.lept4j.Pix;
-import net.sourceforge.lept4j.util.LeptUtils;
 import net.sourceforge.tess4j.ITessAPI;

@Slf4j
@Getter
@RequiredArgsConstructor
-@FieldDefaults(level = AccessLevel.PRIVATE)
+@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
 public class ExtractedOcrImage implements OcrImage {

-    final int pageNumber;
-    final Pix pix;
-    final int originalHeight;
-    final int originalWidth;
-    final int height;
-    final int width;
-    final Matrix ctm;
-    final int numberOnPage;
-
-    @Setter
+    int pageNumber;
+    int numberOnPage;
+    int originalHeight;
+    int originalWidth;
+    Matrix ctm;
+    Pix pix;
+    int height;
+    int width;
    int rotationDegrees;

-    @SneakyThrows
-    public ExtractedOcrImage(int pageNumber, int numberOnPage, BufferedImage bufferedImage, Matrix ctm, int targetDpi, boolean isGray) {
-
-        this.pageNumber = pageNumber;
-        this.numberOnPage = numberOnPage;
-        this.ctm = ctm;
-        this.originalHeight = bufferedImage.getHeight();
-        this.originalWidth = bufferedImage.getWidth();
-        float imageDPI = Math.abs(bufferedImage.getWidth() / (ctm.getScalingFactorX() / 72));
-        this.pix = binarize(bufferedImage, imageDPI, targetDpi, isGray);
-        this.height = pix.h;
-        this.width = pix.w;
-    }
-
-
-    @SneakyThrows
-    private Pix binarize(BufferedImage image, float imageDpi, int targetDpi, boolean isGray) {
-
-        setAlphaChannelToWhite(image);
-        Pix grayScale = convertToGrayScale(image, isGray);
-        Pix scaledUp = scaleToTargetDpi(imageDpi, targetDpi, grayScale);
-        Pix despeckled = LeptUtils.despeckle(scaledUp, LeptUtils.SEL_STR3, 3);
-        LeptUtils.disposePix(scaledUp);
-        return despeckled;
-    }
-
-
-    private static Pix scaleToTargetDpi(float imageDpi, int targetDpi, Pix grayScale) {
-
-        Pix scaledUp;
-        float targetFactor = targetDpi / imageDpi;
-
-        if (targetFactor > 3) {
-            scaledUp = Leptonica1.pixScaleGray4xLI(grayScale);
-            LeptUtils.disposePix(grayScale);
-        } else if (targetFactor > 1.9) {
-            scaledUp = Leptonica1.pixScaleGray2xLI(grayScale);
-            LeptUtils.disposePix(grayScale);
-        } else {
-            scaledUp = grayScale;
-        }
-        return scaledUp;
-    }
-
-
-    private static Pix convertToGrayScale(BufferedImage image, boolean isGray) throws IOException {
-
-        Pix pix = LeptUtils.convertImageToPix(image);
-        Pix grayScale;
-        if (isGray) {
-            grayScale = pix;
-        } else {
-            grayScale = Leptonica1.pixConvertRGBToGrayFast(pix);
-            LeptUtils.disposePix(pix);
-        }
-        return grayScale;
-    }
-
-
-    private static void setAlphaChannelToWhite(BufferedImage image) {
-
-        if (image.getTransparency() == Transparency.TRANSLUCENT) {
-            // NOTE: For BITMASK images, the color model is likely IndexColorModel,
-            // and this model will contain the "real" color of the transparent parts
-            // which is likely a better fit than unconditionally setting it to white.
-
-            // Fill background  with white
-            Graphics2D graphics = image.createGraphics();
-            try {
-                graphics.setComposite(AlphaComposite.DstOver); // Set composite rules to paint "behind"
-                graphics.setPaint(Color.WHITE);
-                graphics.fillRect(0, 0, image.getWidth(), image.getHeight());
-            } finally {
-                graphics.dispose();
-            }
-        }
-    }
-

    @Override
    public AffineTransform getImageCTM() {
@@ -140,11 +58,4 @@ public class ExtractedOcrImage implements OcrImage {
        return affineTransform;
    }

-
-    @Override
-    public int getOptimalPageSegmentationMode() {
-
-        return ITessAPI.TessPageSegMode.PSM_SINGLE_BLOCK;
-    }
-
 }
@@ -2,12 +2,16 @@ package com.knecon.fforesight.service.ocr.processor.model;

 import java.awt.geom.AffineTransform;
 import java.awt.geom.Point2D;
+import java.awt.image.BufferedImage;

+import com.knecon.fforesight.service.ocr.processor.service.threads.OCRThread;
 import com.knecon.fforesight.service.ocr.processor.utils.PdfDpiCalculator;

+import lombok.SneakyThrows;
 import net.sourceforge.lept4j.Leptonica1;
 import net.sourceforge.lept4j.Pix;
 import net.sourceforge.lept4j.util.LeptUtils;
+import net.sourceforge.tess4j.ITessAPI;

 public interface OcrImage {

@@ -27,9 +31,19 @@ public interface OcrImage {
    int getNumberOnPage();


+    /**
+     * Retrieves the height of the original image (not necessarily in pdf coordinates).
+     *
+     * @return the height of the image
+     */
    int getHeight();


+    /**
+     * Retrieves the width of the original image (not necessarily in pdf coordinates).
+     *
+     * @return the width of the image
+     */
    int getWidth();


@@ -40,7 +54,7 @@ public interface OcrImage {
     */
    default QuadPoint getImageBounds() {

-        // cannot be solved with a nice rotation matrix, since the after rotating the text coordinates in the image will always start at (0,0) and will therefore always start at (0,0) in the PDF.
+        // cannot be solved with a nice rotation matrix. After rotating the text coordinates in the image will always start at (0,0) and will therefore always start at (0,0) in the PDF.
        // So in order to mimic this behavior we need to start with (0,0) coordinates always.
        if (getRotationDegrees() == 90 || getRotationDegrees() == 270) {
            return new QuadPoint(new Point2D.Double(0, 0), new Point2D.Double(0, getWidth()), new Point2D.Double(getHeight(), getWidth()), new Point2D.Double(getHeight(), 0));
@@ -74,17 +88,13 @@ public interface OcrImage {
     *
     * @return The optimal page segmentation mode.
     */
-    int getOptimalPageSegmentationMode(); // TODO: evaluate if PSM can be dynamically chosen to increase performance
+    default int getOptimalPageSegmentationMode() {

-
-    /**
-     * Sets the rotation degree of the OCR image. The rotation degree specifies the amount of rotation applied to the image.
-     * Currently only quadrant rotations are supported.
-     * Rotated partial images work, due to the CTM present in the pdf working with any rotation.
-     *
-     * @param rotationDegree The rotation degree of the OCR image.
-     */
-    void setRotationDegrees(int rotationDegree);
+        if (getWidth() < 200 || getHeight() < 200) {
+            return ITessAPI.TessPageSegMode.PSM_SINGLE_BLOCK;
+        }
+        return ITessAPI.TessPageSegMode.PSM_AUTO;
+    } // TODO: evaluate if PSM can be dynamically chosen to increase performance


    /**
@@ -95,22 +105,6 @@ public interface OcrImage {
    Pix getPix();


-    /**
-     * Retrieves the rotated image of the OCR image.
-     *
-     * @return The rotated BufferedImage object of the OCR image.
-     */
-    default Pix getRotatedPix() {
-
-        return switch (360 - getRotationDegrees()) {
-            case 90 -> Leptonica1.pixRotateOrth(getPix(), 1);
-            case 180 -> Leptonica1.pixRotateOrth(getPix(), 2);
-            case 270 -> Leptonica1.pixRotateOrth(getPix(), 3);
-            default -> getPix();
-        };
-    }
-
-
    default int getDpi() {

        return PdfDpiCalculator.calculateDpi(getImageBounds(), getImageCTM(), getWidth());
@@ -125,17 +119,6 @@ public interface OcrImage {
    AffineTransform getImageCTM();


-    /**
-     * Retrieves the size (width * height) of the image.
-     *
-     * @return The size of the image.
-     */
-    default int getImageSize() {
-
-        return getHeight() * getWidth();
-    }
-
-
    default void destroyPix() {

        LeptUtils.disposePix(getPix());
@@ -7,27 +7,17 @@ import com.knecon.fforesight.service.ocr.processor.service.HOcrPageParser;

 import io.github.karols.hocr4j.Word;

-public record OcrResult(Image image, String hOcrPageAbsolutePath) {
+public record OcrResult(OcrImage image, String tesseractOutputFilePath) {

    public static OcrResult create(OcrImage image, String tesseractResult) {

-        return new OcrResult(Image.fromOcrImage(image), tesseractResult);
+        return new OcrResult(image, tesseractResult);
    }


    public List<Word> getAllWords() {

-        return HOcrPageParser.extractHocrPage(hOcrPageAbsolutePath).getAllWords();
-    }
-
-
-    public record Image(Integer pageNumber, AffineTransform ctm, QuadPoint position) {
-
-        public static Image fromOcrImage(OcrImage image) {
-
-            return new Image(image.getPageNumber(), image.getImageCTM(), image.getImageCoordinatesInInitialUserSpace());
-        }
-
+        return HOcrPageParser.extractHocrPage(tesseractOutputFilePath).getAllWords();
    }

 }
@@ -0,0 +1,35 @@
+package com.knecon.fforesight.service.ocr.processor.model;
+
+import java.util.List;
+import java.util.Map;
+import java.util.stream.Collectors;
+
+import com.knecon.fforesight.service.ocr.processor.model.scriptdetection.FontStyleDetectionModel;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.FontMetricsFactory;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.FontStyle;
+
+public record OcrResultToWrite(List<TextPositionInImage> textPositionInImage, QuadPoint imageBoundingBox) {
+
+    public static OcrResultToWrite fromFontStyleDetectionModel(FontStyleDetectionModel fontStyleDetectionModel) {
+
+        return new OcrResultToWrite(fontStyleDetectionModel.getTextPositionInImages(), fontStyleDetectionModel.getImageBounds());
+    }
+
+
+    public static Map<Integer, List<OcrResultToWrite>> buildOcrResultsToWrite(List<OcrResult> ocrResults, FontMetricsFactory fontMetricsFactory) {
+
+        return ocrResults.stream()
+                .collect(Collectors.groupingBy(ocrResult -> ocrResult.image().getPageNumber()))
+                .entrySet()
+                .stream()
+                .collect(Collectors.toMap(Map.Entry::getKey,
+                        entry -> entry.getValue()
+                                .stream()
+                                .map(ocrResult -> new OcrResultToWrite(ocrResult.getAllWords()
+                                        .stream()
+                                        .filter(word -> !word.isBlank())
+                                        .map(word -> new TextPositionInImage(word, ocrResult.image().getImageCTM(), fontMetricsFactory, FontStyle.REGULAR))
+                                        .toList(), ocrResult.image().getImageCoordinatesInInitialUserSpace()))
+                                .toList()));
+    }
+}
@@ -0,0 +1,12 @@
+package com.knecon.fforesight.service.ocr.processor.model;
+
+import org.apache.pdfbox.pdmodel.PDPage;
+
+public record PageInformation(int height, int width, int number, int rotationDegrees) {
+
+    public static PageInformation fromPDPage(int pageNum, PDPage page) {
+
+        return new PageInformation((int) page.getMediaBox().getHeight(), (int) page.getMediaBox().getWidth(), pageNum, page.getRotation());
+    }
+
+}
@@ -97,4 +97,10 @@ public record QuadPoint(Point2D a, Point2D b, Point2D c, Point2D d) {
                d().getY());
    }

+
+    public double size() {
+
+        return a().distance(b()) * a().distance(d());
+    }
+
 }
@@ -1,5 +1,14 @@
 package com.knecon.fforesight.service.ocr.processor.model;

-public record RenderedPageImageFile(int pageNumber, String absoluteFilePath) {
+import net.sourceforge.lept4j.Leptonica1;
+import net.sourceforge.lept4j.Pix;
+
+public record RenderedPageImageFile(int pageNumber, String absoluteFilePath) implements UnprocessedImage {
+
+    @Override
+    public Pix asPix() {
+
+        return Leptonica1.pixRead(absoluteFilePath);
+    }

 }
@@ -8,6 +8,7 @@ import org.apache.pdfbox.pdmodel.PDPage;

 import lombok.AccessLevel;
 import lombok.Getter;
+import lombok.RequiredArgsConstructor;
 import lombok.Setter;
 import lombok.SneakyThrows;
 import lombok.experimental.FieldDefaults;
@@ -16,36 +17,17 @@ import net.sourceforge.lept4j.Pix;
 import net.sourceforge.tess4j.ITessAPI;

@Getter
-@FieldDefaults(level = AccessLevel.PRIVATE)
+@RequiredArgsConstructor
+@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
 public class RenderedPageOcrImage implements OcrImage {

-    final String absoluteImagePath;
-    final int height;
-    final int width;
-    final PageInformation pageInformation;
-    final Pix pix;
-    @Setter
+    int height;
+    int width;
+    PageInformation pageInformation;
+    Pix pix;
    int rotationDegrees;


-    @SneakyThrows
-    public RenderedPageOcrImage(RenderedPageImageFile renderedPageImageFile, PDDocument document) {
-
-        this.pageInformation = PageInformation.fromPDPage(renderedPageImageFile.pageNumber(), document.getPage(renderedPageImageFile.pageNumber() - 1));
-        this.absoluteImagePath = renderedPageImageFile.absoluteFilePath();
-        this.pix = Leptonica1.pixRead(absoluteImagePath);
-        this.height = getPix().h;
-        this.width = getPix().w;
-    }
-
-
-    @Override
-    public int getOptimalPageSegmentationMode() {
-
-        return ITessAPI.TessPageSegMode.PSM_SINGLE_BLOCK;
-    }
-
-
    @Override
    public AffineTransform getImageCTM() {

@@ -78,17 +60,6 @@ public class RenderedPageOcrImage implements OcrImage {
    }


-    @Override
-    public QuadPoint getImageBounds() {
-
-        if (rotationDegrees == 90 || rotationDegrees == 270) {
-            return new QuadPoint(new Point2D.Double(0, 0), new Point2D.Double(0, width), new Point2D.Double(height, width), new Point2D.Double(height, 0));
-        } else {
-            return new QuadPoint(new Point2D.Double(0, 0), new Point2D.Double(0, height), new Point2D.Double(width, height), new Point2D.Double(width, 0));
-        }
-    }
-
-
    @Override
    public int getPageNumber() {

@@ -107,7 +78,7 @@ public class RenderedPageOcrImage implements OcrImage {

        // PDFBox always returns page height and width based on rotation
        double pageWidth;
-        if (pageInformation.rotationDegrees == 90 || pageInformation.rotationDegrees == 270) {
+        if (pageInformation.rotationDegrees() == 90 || pageInformation.rotationDegrees() == 270) {
            pageWidth = pageInformation.height();
        } else {
            pageWidth = pageInformation.width();
@@ -116,14 +87,4 @@ public class RenderedPageOcrImage implements OcrImage {
        return pageWidth / width;
    }

-
-    private record PageInformation(int height, int width, int number, int rotationDegrees) {
-
-        public static PageInformation fromPDPage(int pageNum, PDPage page) {
-
-            return new PageInformation((int) page.getCropBox().getHeight(), (int) page.getCropBox().getWidth(), pageNum, page.getRotation());
-        }
-
-    }
-
 }
@@ -7,29 +7,35 @@ import org.apache.pdfbox.pdmodel.font.PDFont;
 import org.apache.pdfbox.util.Matrix;

 import com.knecon.fforesight.service.ocr.processor.service.fonts.FontMetricsFactory;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.FontStyle;

-import io.github.karols.hocr4j.Bounds;
 import io.github.karols.hocr4j.Word;
 import lombok.AccessLevel;
 import lombok.Getter;
+import lombok.Setter;
 import lombok.experimental.FieldDefaults;

@Getter
-@FieldDefaults(level = AccessLevel.PRIVATE, makeFinal = true)
+@FieldDefaults(level = AccessLevel.PRIVATE)
 public class TextPositionInImage {

-    QuadPoint position;
-    String text;
-    AffineTransform imageCTM;
+    final QuadPoint position;
+    final String text;
+    final AffineTransform imageCTM;
+
+    @Setter
    FontMetricsFactory fontMetricsFactory;
+    @Setter
+    FontStyle fontStyle;


-    public TextPositionInImage(Word word, AffineTransform imageCTM, FontMetricsFactory fontMetricsFactory) {
+    public TextPositionInImage(Word word, AffineTransform imageCTM, FontMetricsFactory fontMetricsFactory, FontStyle fontStyle) {

        this.position = QuadPoint.fromBounds(word.getBounds());
        this.text = word.getText();
        this.imageCTM = imageCTM;
        this.fontMetricsFactory = fontMetricsFactory;
+        this.fontStyle = fontStyle;
    }


@@ -90,6 +96,13 @@ public class TextPositionInImage {
    }


+    public double getTextHeight() {
+
+        var metrics = fontMetricsFactory.calculateMetrics(text, getTransformedWidth(), getTransformedHeight());
+        return fontMetricsFactory.calculateFontSize(text, getTransformedWidth()) * metrics.getHeightScaling();
+    }
+
+
    public double getHeight() {

        return position.a().distance(position.b());
@@ -0,0 +1,9 @@
+package com.knecon.fforesight.service.ocr.processor.model;
+
+import net.sourceforge.lept4j.Pix;
+
+public interface UnprocessedImage {
+
+    Pix asPix();
+
+}
@@ -0,0 +1,58 @@
+package com.knecon.fforesight.service.ocr.processor.model.scriptdetection;
+
+import java.util.List;
+
+import com.knecon.fforesight.service.ocr.processor.model.OcrImage;
+import com.knecon.fforesight.service.ocr.processor.model.OcrResult;
+import com.knecon.fforesight.service.ocr.processor.model.QuadPoint;
+import com.knecon.fforesight.service.ocr.processor.model.TextPositionInImage;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.FontMetricsFactory;
+import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;
+
+import lombok.AccessLevel;
+import lombok.Getter;
+import lombok.RequiredArgsConstructor;
+import lombok.experimental.FieldDefaults;
+import net.sourceforge.lept4j.Leptonica1;
+import net.sourceforge.lept4j.Pix;
+import net.sourceforge.lept4j.util.LeptUtils;
+
+@Getter
+@RequiredArgsConstructor
+@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
+public final class FontStyleDetectionModel {
+
+    QuadPoint imageBounds;
+    Pix image;
+    List<TextPositionAndWordImage> textPositionsAndWordImages;
+
+
+    public static FontStyleDetectionModel fromOcrResult(OcrResult ocrResult, FontMetricsFactory fontMetricsFactory, OcrServiceSettings settings) {
+
+        var image = Leptonica1.pixRead(ocrResult.tesseractOutputFilePath() + ".tiff");
+        var wordPixes = ocrResult.getAllWords().stream().filter(word -> !word.isBlank()).map(word -> TextPositionAndWordImage.create(ocrResult.image().getImageCTM(), word, image, settings, fontMetricsFactory)).toList();
+
+        return new FontStyleDetectionModel(ocrResult.image().getImageCoordinatesInInitialUserSpace(), image, wordPixes);
+    }
+
+
+    public List<TextPositionInImage> getTextPositionInImages() {
+
+        return textPositionsAndWordImages.stream().map(TextPositionAndWordImage::getTextPositionInImage).toList();
+    }
+
+
+    public List<WordImage> getWordImages() {
+
+        return textPositionsAndWordImages.stream().map(TextPositionAndWordImage::getWordImage).toList();
+    }
+
+
+    public void dispose() {
+
+        LeptUtils.disposePix(image);
+        getWordImages().forEach(WordImage::dispose);
+    }
+
+
+}
@@ -0,0 +1,52 @@
+package com.knecon.fforesight.service.ocr.processor.model.scriptdetection;
+
+import java.awt.geom.AffineTransform;
+import java.util.Objects;
+
+import org.apache.commons.math3.ml.clustering.Clusterable;
+
+import com.knecon.fforesight.service.ocr.processor.model.OcrResult;
+import com.knecon.fforesight.service.ocr.processor.model.TextPositionInImage;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.FontMetricsFactory;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.FontStyle;
+import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;
+
+import io.github.karols.hocr4j.Word;
+import lombok.Getter;
+import net.sourceforge.lept4j.Pix;
+
+@Getter
+public final class TextPositionAndWordImage implements Clusterable {
+
+    private final TextPositionInImage textPositionInImage;
+    private final WordImage wordImage;
+
+
+    public TextPositionAndWordImage(TextPositionInImage textPositionInImage, WordImage wordImage) {
+
+        this.textPositionInImage = textPositionInImage;
+        this.wordImage = wordImage;
+    }
+
+
+    public static TextPositionAndWordImage create(AffineTransform imageCTM, Word word, Pix image, OcrServiceSettings settings, FontMetricsFactory fontMetricsFactory) {
+
+        TextPositionInImage textPositionInImage = new TextPositionInImage(word, imageCTM, fontMetricsFactory, FontStyle.REGULAR);
+        WordImage wordImage = new WordImage(textPositionInImage.getTextHeight(), word, image, settings);
+        return new TextPositionAndWordImage(textPositionInImage, wordImage);
+    }
+
+
+    @Override
+    public double[] getPoint() {
+
+        return wordImage.getPoint();
+    }
+
+
+    public double getTextHeight() {
+
+        return wordImage.getTextHeight();
+    }
+
+}
@@ -0,0 +1,71 @@
+package com.knecon.fforesight.service.ocr.processor.model.scriptdetection;
+
+import org.apache.commons.math3.ml.clustering.Clusterable;
+
+import com.knecon.fforesight.service.ocr.processor.service.scriptdetection.StrokeWidthCalculator;
+import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;
+import com.knecon.fforesight.service.ocr.processor.utils.ImageProcessingUtils;
+
+import io.github.karols.hocr4j.Word;
+import lombok.AccessLevel;
+import lombok.Getter;
+import lombok.RequiredArgsConstructor;
+import lombok.experimental.FieldDefaults;
+import net.sourceforge.lept4j.Box;
+import net.sourceforge.lept4j.Leptonica1;
+import net.sourceforge.lept4j.Pix;
+import net.sourceforge.lept4j.util.LeptUtils;
+
+@Getter
+@RequiredArgsConstructor
+@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
+public class WordImage implements Clusterable {
+
+    Pix image;
+    String text;
+    double textHeight;
+    OcrServiceSettings settings;
+
+
+    public WordImage(double textHeight, Word word, Pix originalImage, OcrServiceSettings settings) {
+
+        Box box = new Box(word.getBounds().getLeft(), word.getBounds().getTop(), word.getBounds().getWidth(), word.getBounds().getHeight(), 1);
+        this.image = Leptonica1.pixClipRectangle(originalImage, box, null);
+        box.clear();
+        this.text = word.getText();
+        this.textHeight = textHeight;
+        this.settings = settings;
+    }
+
+
+    public boolean hasLargerStrokeWidth(double strokeWidth) {
+
+        int roundedStrokeWidth = (int) Math.round(strokeWidth);
+        double roundingError = (roundedStrokeWidth - strokeWidth) / strokeWidth;
+
+        // add 1 to open a bit bigger than the estimated regular stroke width
+        Pix openedPix = Leptonica1.pixOpenBrick(null, image, roundedStrokeWidth + 1, roundedStrokeWidth + 1);
+
+        double openedPixelDensity = ImageProcessingUtils.calculatePixelDensity(openedPix);
+
+        double pixelDensity = ImageProcessingUtils.calculatePixelDensity(image);
+
+        LeptUtils.disposePix(openedPix);
+
+        return (openedPixelDensity * (1 + roundingError)) / pixelDensity > (settings.getBoldThreshold());
+    }
+
+
+    @Override
+    public double[] getPoint() {
+
+        return new double[]{textHeight};
+    }
+
+
+    public void dispose() {
+
+        LeptUtils.disposePix(image);
+    }
+
+}
@@ -3,19 +3,21 @@ package com.knecon.fforesight.service.ocr.processor.service;
 import java.io.InputStream;
 import java.nio.file.Path;
 import java.util.ArrayList;
-import java.util.Collections;
+import java.util.HashMap;
 import java.util.LinkedList;
 import java.util.List;
+import java.util.Map;
 import java.util.concurrent.BlockingQueue;
+import java.util.concurrent.LinkedBlockingDeque;
 import java.util.stream.Collectors;

 import org.apache.pdfbox.pdmodel.PDDocument;
 import org.springframework.stereotype.Service;

-import com.knecon.fforesight.service.ocr.processor.model.OcrImage;
 import com.knecon.fforesight.service.ocr.processor.model.RenderedPageImageFile;
-import com.knecon.fforesight.service.ocr.processor.model.RenderedPageOcrImage;
-import com.knecon.fforesight.service.ocr.processor.service.threads.ProcessIOLogger;
+import com.knecon.fforesight.service.ocr.processor.model.UnprocessedImage;
+import com.knecon.fforesight.service.ocr.processor.service.threads.BlockingQueueFiller;
+import com.knecon.fforesight.service.ocr.processor.service.threads.GhostScriptOutputHandler;
 import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;
 import com.knecon.fforesight.service.ocr.processor.utils.ListSplittingUtils;

@@ -42,17 +44,19 @@ public class GhostScriptService {
                                                        String documentAbsolutePath,
                                                        Path tmpImageDir,
                                                        PDDocument document,
-                                                        BlockingQueue<OcrImage> imageOutputQueue,
+                                                        BlockingQueue<UnprocessedImage> imageProcessingQueue,
                                                        Statistics stats) {

+        BlockingQueue<RenderedPageImageFile> imageFileCollectorQueue = new LinkedBlockingDeque<>();
+        BlockingQueueFiller asyncTransferThread = new BlockingQueueFiller(imageFileCollectorQueue, imageProcessingQueue);
+        asyncTransferThread.start();
        int numOfProcesses = Math.min(settings.getGsProcessCount(), stitchedPageNumbers.size());

        List<List<ProcessInfo>> processInfoBatches = buildSubListForEachProcess(stitchedPageNumbers,
                numOfProcesses,
-                2 * settings.getOcrThreadCount()); // use 2 times the thread count as batch size, such that GS generates the rendered pages as needed by the OCR Threads
+                256 * numOfProcesses); // GS has a limit on how many pageIndices per call are possible, so we limit it to 256 pages per process
        for (int batchIdx = 0; batchIdx < processInfoBatches.size(); batchIdx++) {
            long timestamp = System.currentTimeMillis();
-            List<RenderedPageImageFile> renderedPageImageFiles = Collections.synchronizedList(new LinkedList<>());
            List<ProcessInfo> processInfos = processInfoBatches.get(batchIdx);

            log.info("Batch {}: Running {} gs processes with ({}) pages each",
@@ -63,9 +67,9 @@ public class GhostScriptService {
            int finalBatchIdx = batchIdx;
            List<Process> processes = processInfos.stream()
                    .parallel()
-                    .map(info -> buildCmdArgs(info.processIdx(), finalBatchIdx, info.stitchedPageNumbers(), tmpImageDir, documentAbsolutePath, renderedPageImageFiles))
-                    .peek(s -> log.debug(String.join(" ", s)))
-                    .map(this::executeProcess)
+                    .map(info -> buildCmdArgs(info.processIdx(), finalBatchIdx, info.stitchedPageNumbers(), tmpImageDir, documentAbsolutePath))
+                    .peek(s -> log.debug(String.join(" ", s.cmdArgs())))
+                    .map(processInfo -> executeProcess(processInfo, imageFileCollectorQueue))
                    .toList();

            List<Integer> processExitCodes = new LinkedList<>();
@@ -73,14 +77,9 @@ public class GhostScriptService {
                processExitCodes.add(process.waitFor());
            }
            stats.increasePDF2ImgDuration(System.currentTimeMillis() - timestamp);
-
            log.info("Batch {}: Ghostscript processes finished with exit codes " + processExitCodes, batchIdx);
-            for (RenderedPageImageFile renderedPageImageFile : renderedPageImageFiles) {
-                OcrImage image = new RenderedPageOcrImage(renderedPageImageFile, document);
-                imageOutputQueue.put(image);
-            }
-
        }
+        asyncTransferThread.setAllImagesQueued(true);
    }


@@ -107,20 +106,28 @@ public class GhostScriptService {


    @SneakyThrows
-    private String[] buildCmdArgs(Integer processIdx,
-                                  Integer batchIdx,
-                                  List<Integer> stitchedImagePageIndices,
-                                  Path outputDir,
-                                  String documentAbsolutePath,
-                                  List<RenderedPageImageFile> fullPageImages) {
+    private ProcessCmdsAndRenderedImageFiles buildCmdArgs(Integer processIdx,
+                                                          Integer batchIdx,
+                                                          List<Integer> stitchedImagePageIndices,
+                                                          Path outputDir,
+                                                          String documentAbsolutePath) {

        String imagePathFormat = outputDir.resolve("output_" + processIdx + "_" + batchIdx + ".%04d" + FORMAT).toFile().toString();

+        Map<Integer, RenderedPageImageFile> fullPageImages = new HashMap<>();
        for (int i = 0; i < stitchedImagePageIndices.size(); i++) {
            Integer pageNumber = stitchedImagePageIndices.get(i);
-            fullPageImages.add(new RenderedPageImageFile(pageNumber, String.format(imagePathFormat, i + 1)));
+            fullPageImages.put(pageNumber, new RenderedPageImageFile(pageNumber, String.format(imagePathFormat, i + 1)));
        }

+        String[] cmdArgs = buildCmdArgs(stitchedImagePageIndices, documentAbsolutePath, imagePathFormat);
+
+        return new ProcessCmdsAndRenderedImageFiles(cmdArgs, fullPageImages);
+    }
+
+
+    private String[] buildCmdArgs(List<Integer> stitchedImagePageIndices, String documentAbsolutePath, String imagePathFormat) {
+
        StringBuilder sPageList = new StringBuilder();
        int i = 1;
        for (Integer integer : stitchedImagePageIndices) {
@@ -131,18 +138,19 @@ public class GhostScriptService {
            i++;
        }

-        return new String[]{"gs", "-dNOPAUSE", "-sDEVICE=" + DEVICE, "-r" + settings.getDpi(), "-sPageList=" + sPageList, "-sOutputFile=" + imagePathFormat, documentAbsolutePath, "-c", "quit"};
+        String[] cmdArgs = new String[]{"gs", "-dNOPAUSE", "-sDEVICE=" + DEVICE, "-r" + settings.getDpi(), "-sPageList=" + sPageList, "-sOutputFile=" + imagePathFormat, documentAbsolutePath, "-c", "quit"};
+        return cmdArgs;
    }


    @SneakyThrows
-    private Process executeProcess(String[] cmdArgs) {
+    private Process executeProcess(ProcessCmdsAndRenderedImageFiles processInfo, BlockingQueue<RenderedPageImageFile> imageFileCollectorQueue) {

-        Process p = Runtime.getRuntime().exec(cmdArgs);
+        Process p = Runtime.getRuntime().exec(processInfo.cmdArgs());
        InputStream stdOut = p.getInputStream();
-        ProcessIOLogger stdOutLogger = new ProcessIOLogger(stdOut, "GS", ProcessIOLogger.Type.STD_OUT);
+        GhostScriptOutputHandler stdOutLogger = GhostScriptOutputHandler.stdOut(stdOut, processInfo.renderedPageImageFiles(), imageFileCollectorQueue);
        InputStream stdError = p.getErrorStream();
-        ProcessIOLogger stdErrorLogger = new ProcessIOLogger(stdError, "GS", ProcessIOLogger.Type.ERROR);
+        GhostScriptOutputHandler stdErrorLogger = GhostScriptOutputHandler.errorHandler(stdError);

        stdOutLogger.start();
        stdErrorLogger.start();
@@ -150,6 +158,10 @@ public class GhostScriptService {
    }


+    private record ProcessCmdsAndRenderedImageFiles(String[] cmdArgs, Map<Integer, RenderedPageImageFile> renderedPageImageFiles) {
+
+    }
+
    private record ProcessInfo(Integer processIdx, List<Integer> stitchedPageNumbers) {

    }
@@ -1,7 +1,6 @@
 package com.knecon.fforesight.service.ocr.processor.service;

-import java.awt.Graphics;
-import java.awt.image.BufferedImage;
+import java.awt.geom.Rectangle2D;
 import java.io.IOException;
 import java.util.LinkedList;
 import java.util.List;
@@ -18,13 +17,12 @@ import org.apache.pdfbox.cos.COSBase;
 import org.apache.pdfbox.cos.COSName;
 import org.apache.pdfbox.pdmodel.PDPage;
 import org.apache.pdfbox.pdmodel.graphics.PDXObject;
-import org.apache.pdfbox.pdmodel.graphics.color.PDDeviceGray;
-import org.apache.pdfbox.pdmodel.graphics.color.PDDeviceRGB;
 import org.apache.pdfbox.pdmodel.graphics.form.PDFormXObject;
 import org.apache.pdfbox.pdmodel.graphics.image.PDImageXObject;
 import org.apache.pdfbox.util.Matrix;

-import com.knecon.fforesight.service.ocr.processor.model.ExtractedOcrImage;
+import com.knecon.fforesight.service.ocr.processor.model.ExtractedImage;
+import com.knecon.fforesight.service.ocr.processor.model.QuadPoint;
 import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;

 import lombok.Getter;
@@ -33,8 +31,7 @@ import lombok.SneakyThrows;
@Getter
 public class ImageStreamEngine extends PDFStreamEngine {

-    private ExtractedOcrImage currentImageOnPage;
-    private List<ExtractedOcrImage> imagesOnCurrentPage;
+    private List<ExtractedImage> imagesOnCurrentPage;
    private OcrServiceSettings settings;
    private int pageNum;

@@ -69,23 +66,15 @@ public class ImageStreamEngine extends PDFStreamEngine {
                }

                Matrix imageCTM = getGraphicsState().getCurrentTransformationMatrix();
-                if (imageXObject.getColorSpace() instanceof PDDeviceRGB) {
-                    BufferedImage image = imageXObject.getImage();
-                    this.currentImageOnPage = new ExtractedOcrImage(pageNum, imagesOnCurrentPage.size(), image, imageCTM, settings.getDpi(), false);
-                } else if (imageXObject.getColorSpace() instanceof PDDeviceGray) {
-                    BufferedImage image = imageXObject.getImage();
-                    this.currentImageOnPage = new ExtractedOcrImage(pageNum, imagesOnCurrentPage.size(), image, imageCTM, settings.getDpi(), true);
-                } else {
-                    BufferedImage pdfImage = imageXObject.getImage();
-                    BufferedImage image = new BufferedImage(pdfImage.getWidth(), pdfImage.getHeight(),
-                            BufferedImage.TYPE_BYTE_GRAY);
-                    Graphics g = image.getGraphics();
-                    g.drawImage(pdfImage, 0, 0, null);
-                    g.dispose();
-                    this.currentImageOnPage = new ExtractedOcrImage(pageNum, imagesOnCurrentPage.size(), image, imageCTM, settings.getDpi(), true);
-                }
-                this.imagesOnCurrentPage.add(this.currentImageOnPage);
-                //imagesOnPages.add(this.currentImageOnPage);
+                this.imagesOnCurrentPage.add(new ExtractedImage(pageNum,
+                    QuadPoint.fromRectangle2D(new Rectangle2D.Double(0, 0, imageXObject.getWidth(), imageXObject.getHeight())),
+                        imageXObject.getHeight(),
+                        imageXObject.getWidth(),
+                        imageXObject.getImage(),
+                        imageCTM,
+                        imagesOnCurrentPage.size(),
+                        imageXObject.getColorSpace()));
+
            } else if (xobject instanceof PDFormXObject) {
                PDFormXObject form = (PDFormXObject) xobject;
                showForm(form);
@@ -9,6 +9,7 @@ import java.io.OutputStream;
 import java.nio.file.Path;
 import java.util.LinkedList;
 import java.util.List;
+import java.util.Map;
 import java.util.concurrent.ArrayBlockingQueue;
 import java.util.concurrent.BlockingQueue;
 import java.util.stream.IntStream;
@@ -20,8 +21,10 @@ import org.springframework.util.FileSystemUtils;

 import com.iqser.red.pdftronlogic.commons.InvisibleElementRemovalService;
 import com.iqser.red.pdftronlogic.commons.WatermarkRemovalService;
+import com.knecon.fforesight.service.ocr.processor.model.OcrResultToWrite;
 import com.knecon.fforesight.service.ocr.processor.model.OcrImage;
 import com.knecon.fforesight.service.ocr.processor.model.OcrResult;
+import com.knecon.fforesight.service.ocr.processor.service.scriptdetection.FontStyleDetector;
 import com.knecon.fforesight.service.ocr.processor.service.threads.OCRThread;
 import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;

@@ -44,6 +47,7 @@ public class OCRService {
    InvisibleElementRemovalService invisibleElementRemovalService;
    OcrResultWriter ocrResultWriter;
    GhostScriptService ghostScriptService;
+    FontStyleDetector boldDetector;


    /**
@@ -107,7 +111,7 @@ public class OCRService {
            int numberOfOcrThreads = Math.min(settings.getOcrThreadCount(), document.getNumberOfPages());
            stats = new Statistics(numberOfExtractThreads, numberOfOcrThreads);

-            BlockingQueue<OcrImage> ocrImageQueue = new ArrayBlockingQueue<>(numberOfOcrThreads);
+            BlockingQueue<OcrImage> ocrImageQueue = new ArrayBlockingQueue<>((int) (1.5 * numberOfOcrThreads));

            OcrImageFactory ocrImageFactory = new OcrImageFactory(document,
                    documentFile,
@@ -128,16 +132,21 @@ public class OCRService {
                    .toList();
            log.info("Started {} OCR consumer threads, listening for images on the queue", ocrThreads.size());
            ocrImageFactory.join();
-            log.info("Extracted all images, interrupting ocr threads");
+            log.info("Processed all images, interrupting ocr threads");

            ocrThreads.forEach(Thread::interrupt);
            for (OCRThread ocrThread : ocrThreads) {
                ocrThread.join();
            }

-            log.info("OCR processing has finished, writing results");
+            log.info("Tesseract OCR has finished for file {} and dossier {}", fileId, dossierId);
+
            timestamp = System.currentTimeMillis();
-            var dictionariesToUpdate = ocrResultWriter.drawOcrResultsToPdf(document, ocrResults);
+            Map<Integer, List<OcrResultToWrite>> imageWithTextPositionsPerPage = boldDetector.detectBold(ocrResults, document);
+            stats.increaseFontStyleDetectionDuration(System.currentTimeMillis() - timestamp);
+
+            timestamp = System.currentTimeMillis();
+            var dictionariesToUpdate = ocrResultWriter.drawOcrResultsToPdf(document, imageWithTextPositionsPerPage);
            log.info("Saving document");
            document.saveIncremental(out, dictionariesToUpdate);
            stats.increaseWritingTextDuration(System.currentTimeMillis() - timestamp);
@@ -6,13 +6,17 @@ import java.util.ArrayList;
 import java.util.Collections;
 import java.util.LinkedList;
 import java.util.List;
+import java.util.concurrent.ArrayBlockingQueue;
 import java.util.concurrent.BlockingQueue;
 import java.util.stream.Collectors;

 import org.apache.pdfbox.pdmodel.PDDocument;

+import com.knecon.fforesight.service.ocr.processor.model.ExtractedImage;
 import com.knecon.fforesight.service.ocr.processor.model.OcrImage;
+import com.knecon.fforesight.service.ocr.processor.model.UnprocessedImage;
 import com.knecon.fforesight.service.ocr.processor.service.threads.ImageExtractionThread;
+import com.knecon.fforesight.service.ocr.processor.service.threads.ImageProcessingThread;
 import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;
 import com.knecon.fforesight.service.ocr.processor.utils.ListSplittingUtils;

@@ -29,6 +33,8 @@ public class OcrImageFactory {
    File documentFile;
    Path tmpImageDir;
    GhostScriptService ghostScriptService;
+    BlockingQueue<UnprocessedImage> imageProcessingQueue;
+    ImageProcessingThread imageProcessingThread;
    BlockingQueue<OcrImage> imageOutputQueue;
    List<ImageExtractionThread> imageExtractionThreads;
    List<Integer> stitchedPageNumbers;
@@ -40,7 +46,7 @@ public class OcrImageFactory {
                           Path tmpImageDir,
                           int numberOfThreads,
                           GhostScriptService ghostScriptService,
-                           BlockingQueue<OcrImage> imageOutputQueue,
+                           BlockingQueue<OcrImage> imageOcrQueue,
                           OcrProgressLogger logger,
                           OcrServiceSettings settings,
                           Statistics stats) {
@@ -49,7 +55,8 @@ public class OcrImageFactory {
        this.documentFile = documentFile;
        this.tmpImageDir = tmpImageDir;
        this.ghostScriptService = ghostScriptService;
-        this.imageOutputQueue = imageOutputQueue;
+        this.imageOutputQueue = imageOcrQueue;
+        this.imageProcessingQueue = new ArrayBlockingQueue<>(imageOcrQueue.remainingCapacity());
        this.stitchedPageNumbers = Collections.synchronizedList(new LinkedList<>());
        this.stats = stats;

@@ -57,8 +64,10 @@ public class OcrImageFactory {

        List<List<Integer>> balancedPageNumbers = ListSplittingUtils.buildBalancedContinuousSublist(document.getNumberOfPages(), numberOfThreads);
        for (int i = 0; i < balancedPageNumbers.size(); i++) {
-            imageExtractionThreads.add(new ImageExtractionThread(i, balancedPageNumbers.get(i), documentFile, logger, stats, settings, imageOutputQueue, stitchedPageNumbers));
+            imageExtractionThreads.add(new ImageExtractionThread(i, balancedPageNumbers.get(i), documentFile, logger, stats, settings, imageProcessingQueue, stitchedPageNumbers));
        }
+        this.imageProcessingThread = new ImageProcessingThread(imageProcessingQueue, imageOcrQueue, stats, settings, document);
+
        log.info("Started {} image extraction threads, with ({}) pages each",
                imageExtractionThreads.size(),
                imageExtractionThreads.stream().map(ImageExtractionThread::getPageIndices).map(List::size).map(String::valueOf).collect(Collectors.joining(", ")));
@@ -70,6 +79,8 @@ public class OcrImageFactory {
        for (ImageExtractionThread imageExtractionThread : imageExtractionThreads) {
            imageExtractionThread.start();
        }
+        imageProcessingThread.start();
+
    }


@@ -79,11 +90,16 @@ public class OcrImageFactory {
        for (ImageExtractionThread imageExtractionThread : imageExtractionThreads) {
            imageExtractionThread.join();
        }
-        if (stitchedPageNumbers.isEmpty()) {
-            return;
+
+        if (!stitchedPageNumbers.isEmpty()) {
+            ghostScriptService.renderPagesAsImagesBatchedAndAddToQueue(stitchedPageNumbers, documentFile.toString(), tmpImageDir, document, imageProcessingQueue, stats);
        }

-        ghostScriptService.renderPagesAsImagesBatchedAndAddToQueue(stitchedPageNumbers, documentFile.toString(), tmpImageDir, document, imageOutputQueue, stats);
+        imageProcessingThread.setAllImagesExtracted(true);
+        imageProcessingThread.interrupt();
+
+        imageProcessingThread.join();
+
    }

 }
@@ -2,11 +2,11 @@ package com.knecon.fforesight.service.ocr.processor.service;

 import java.awt.Color;
 import java.awt.geom.Point2D;
+import java.util.Collection;
 import java.util.HashSet;
 import java.util.List;
 import java.util.Map;
 import java.util.Set;
-import java.util.stream.Collectors;

 import org.apache.pdfbox.cos.COSDictionary;
 import org.apache.pdfbox.cos.COSName;
@@ -20,11 +20,9 @@ import org.apache.pdfbox.pdmodel.graphics.optionalcontent.PDOptionalContentPrope
 import org.apache.pdfbox.pdmodel.graphics.state.RenderingMode;
 import org.springframework.stereotype.Service;

-import com.knecon.fforesight.service.ocr.processor.model.OcrResult;
+import com.knecon.fforesight.service.ocr.processor.model.OcrResultToWrite;
 import com.knecon.fforesight.service.ocr.processor.model.QuadPoint;
 import com.knecon.fforesight.service.ocr.processor.model.TextPositionInImage;
-import com.knecon.fforesight.service.ocr.processor.service.fonts.FontMetricsFactory;
-import com.knecon.fforesight.service.ocr.processor.service.fonts.Type0FontMetricsFactory;
 import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;

 import lombok.AccessLevel;
@@ -44,19 +42,17 @@ public class OcrResultWriter {


    @SneakyThrows
-    public Set<COSDictionary> drawOcrResultsToPdf(PDDocument document, List<OcrResult> ocrResults) {
+    public Set<COSDictionary> drawOcrResultsToPdf(PDDocument document, Map<Integer, List<OcrResultToWrite>> imagesWithResultsPerPage) {

-        FontMetricsFactory fontMetricsFactory = new Type0FontMetricsFactory(document);
        Set<COSDictionary> dictionariesToUpdate = new HashSet<>();
-        Map<Integer, List<OcrResult>> resultsPerPage = ocrResults.stream().collect(Collectors.groupingBy(result -> result.image().pageNumber()));
-        resultsPerPage.keySet().forEach(pageNumber -> drawResultsPerPage(document, pageNumber, resultsPerPage, dictionariesToUpdate, fontMetricsFactory));
+        imagesWithResultsPerPage.keySet().forEach(pageNumber -> drawResultsPerPage(document, pageNumber, imagesWithResultsPerPage.get(pageNumber), dictionariesToUpdate));
        dictionariesToUpdate.add(document.getDocumentInformation().getCOSObject());
        return dictionariesToUpdate;
    }


    @SneakyThrows
-    private void drawResultsPerPage(PDDocument document, Integer pageNumber, Map<Integer, List<OcrResult>> resultsPerPage, Set<COSDictionary> dictionariesToUpdate, FontMetricsFactory fontMetricsFactory) {
+    private void drawResultsPerPage(PDDocument document, Integer pageNumber, List<OcrResultToWrite> ocrResultToWrite, Set<COSDictionary> dictionariesToUpdate) {

        var pdPage = document.getPage(pageNumber - 1);

@@ -69,7 +65,7 @@ public class OcrResultWriter {

        escapeContentStreams(document, pdPage);

-        List<TextPositionInImage> words = buildTextPositionsOnPage(pageNumber, resultsPerPage, fontMetricsFactory);
+        List<TextPositionInImage> words = ocrResultToWrite.stream().map(OcrResultToWrite::textPositionInImage).flatMap(Collection::stream).toList();
        try (var contentStream = new PDPageContentStream(document, pdPage, PDPageContentStream.AppendMode.APPEND, true)) {

            // write invisible ocr text inside tagged content
@@ -86,7 +82,6 @@ public class OcrResultWriter {
                // write visible ocr text inside optional group
                contentStream.beginMarkedContent(COSName.OC, textDebugLayer);
                contentStream.saveGraphicsState();
-                contentStream.setNonStrokingColor(Color.BLUE);
                words.forEach(word -> drawVisibleWord(word, contentStream));
                contentStream.restoreGraphicsState();
                contentStream.endMarkedContent();
@@ -94,7 +89,9 @@ public class OcrResultWriter {
                // write word bounding boxes (tesseract output) inside optional group
                contentStream.beginMarkedContent(COSName.OC, bBoxDebugLayer);
                contentStream.saveGraphicsState();
-                resultsPerPage.get(pageNumber).stream().map(OcrResult::image).forEach(image -> drawGrid(contentStream, image.position()));
+                ocrResultToWrite.stream()
+                        .map(OcrResultToWrite::imageBoundingBox)
+                        .forEach(imagePosition -> drawGrid(contentStream, imagePosition));
                words.stream().map(TextPositionInImage::getTransformedTextBBox).forEach(word -> drawRectangle(contentStream, word));
                contentStream.restoreGraphicsState();
                contentStream.endMarkedContent();
@@ -105,15 +102,6 @@ public class OcrResultWriter {
    }


-    private static List<TextPositionInImage> buildTextPositionsOnPage(Integer pageNumber, Map<Integer, List<OcrResult>> resultsPerPage, FontMetricsFactory fontMetricsFactory) {
-
-        return resultsPerPage.get(pageNumber)
-                .stream()
-                .flatMap(result -> result.getAllWords().stream().filter(word -> !word.isBlank()).map(word -> new TextPositionInImage(word, result.image().ctm(), fontMetricsFactory)))
-                .toList();
-    }
-
-
    @SneakyThrows
    private static void escapeContentStreams(PDDocument document, PDPage pdPage) {
        // We need to append to the contentstream, otherwise the content could be overlapped by images
@@ -196,6 +184,11 @@ public class OcrResultWriter {
    private void drawWord(TextPositionInImage position, PDPageContentStream contentStream, RenderingMode renderingMode) {

        try {
+            contentStream.setNonStrokingColor(switch (position.getFontStyle()) {
+                case BOLD -> Color.RED;
+                case ITALIC -> Color.GREEN;
+                default -> Color.BLUE;
+            });
            contentStream.beginText();
            contentStream.setRenderingMode(renderingMode);
            contentStream.setFont(position.getFont(), (float) position.getFontSize());
@@ -15,14 +15,18 @@ public class Statistics {
    List<Long> tesseractDuration;
    AtomicLong pdf2ImgDuration;
    AtomicLong writingTextDuration;
+    AtomicLong imageProcessingDuration;
+    AtomicLong fontStyleDetectionDuration;


    public Statistics(int numberOfExtractThreads, int numberOfOcrThreads) {

        this.imageExtraction = Collections.synchronizedList(new ArrayList<>(Collections.nCopies(numberOfExtractThreads, 0L)));
        this.tesseractDuration = Collections.synchronizedList(new ArrayList<>(Collections.nCopies(numberOfOcrThreads, 0L)));
+        this.fontStyleDetectionDuration = new AtomicLong(0);
        this.pdf2ImgDuration = new AtomicLong(0);
        this.writingTextDuration = new AtomicLong(0);
+        this.imageProcessingDuration = new AtomicLong(0);
    }


@@ -32,6 +36,12 @@ public class Statistics {
    }


+    public void increaseImageProcessing(long duration) {
+
+        imageProcessingDuration.addAndGet(duration);
+    }
+
+
    public void increaseTesseractDuration(int threadId, long duration) {

        tesseractDuration.set(threadId, tesseractDuration.get(threadId) + duration);
@@ -49,19 +59,27 @@ public class Statistics {
        writingTextDuration.addAndGet(duration);
    }

+    public void increaseFontStyleDetectionDuration(long duration) {
+
+        fontStyleDetectionDuration.addAndGet(duration);
+    }
+

    @Override
    public String toString() {

-        return String.format("imageExtraction: mean %.2f s, max %.2f s, min %.2f, tesseract: mean %.2f s, max %.2f s, min %.2f, PDF2Img=%.2f s, writingText=%.2f s",
+        return String.format(
+                "imageExtraction: mean %.2f s, max %.2f s, min %.2f, tesseract: mean %.2f s, max %.2f s, min %.2f, ImageProcessing=%.2f s, PDF2Img=%.2f s, writingText=%.2f s, FontstyleDetection=%.2f s",
                ((float) imageExtraction.stream().mapToLong(Long::longValue).average().orElse(0) / 1000),
                ((float) imageExtraction.stream().mapToLong(Long::longValue).max().orElse(0) / 1000),
                ((float) imageExtraction.stream().mapToLong(Long::longValue).min().orElse(0) / 1000),
                ((float) tesseractDuration.stream().mapToLong(Long::longValue).average().orElse(0) / 1000),
                ((float) tesseractDuration.stream().mapToLong(Long::longValue).max().orElse(0) / 1000),
                ((float) tesseractDuration.stream().mapToLong(Long::longValue).min().orElse(0) / 1000),
+                (float) imageProcessingDuration.get() / 1000,
                (float) pdf2ImgDuration.get() / 1000,
-                (float) writingTextDuration.get() / 1000);
+                (float) writingTextDuration.get() / 1000,
+                (float) fontStyleDetectionDuration.get() / 1000);
    }

 }
@@ -36,6 +36,7 @@ public interface FontMetricsFactory {

    PDFont getFont();

+
    HeightAndDescent calculateHeightAndDescent(String text);

 }
@@ -0,0 +1,5 @@
+package com.knecon.fforesight.service.ocr.processor.service.fonts;
+
+public enum FontStyle {
+    REGULAR, BOLD, ITALIC
+}
@@ -1,6 +1,9 @@
 package com.knecon.fforesight.service.ocr.processor.service.fonts;

 import java.io.ByteArrayInputStream;
+import java.util.Collections;
+import java.util.List;
+import java.util.Set;

 import org.apache.fontbox.ttf.GlyphData;
 import org.apache.fontbox.ttf.TTFParser;
@@ -12,22 +15,41 @@ import org.apache.pdfbox.pdmodel.font.PDType0Font;

 import com.knecon.fforesight.service.ocr.processor.model.HeightAndDescent;

+import lombok.RequiredArgsConstructor;
 import lombok.SneakyThrows;
 import lombok.extern.slf4j.Slf4j;
+import software.amazon.awssdk.services.s3.endpoints.internal.Value;

@Slf4j
+@RequiredArgsConstructor
 public class Type0FontMetricsFactory implements FontMetricsFactory {

    private final PDType0Font type0Font;
    private final TrueTypeFont trueTypeFont;

+    // for this specific font back-/forward-slashes have a lot of descent screwing up the font size and therefore bold detection. So if we find such a character we ignore its descent.
+    private static final Set<Integer> slashGlyphIds = Set.of(18, 63);
+
+
+    public static Type0FontMetricsFactory regular(PDDocument document) {
+
+        return createFromResource("fonts/cmu-regular.ttf", document);
+    }
+
+
+    public static Type0FontMetricsFactory bold(PDDocument document) {
+
+        return createFromResource("fonts/cmu-bold.ttf", document);
+    }
+

    @SneakyThrows
-    public Type0FontMetricsFactory(PDDocument document) {
+    private static Type0FontMetricsFactory createFromResource(String resourcePath, PDDocument document) {

-        try (var in = Thread.currentThread().getContextClassLoader().getResourceAsStream("fonts/cmu-regular.ttf"); var buffer = new RandomAccessReadBuffer(in)) {
-            this.trueTypeFont = new TTFParser().parse(buffer); // since Type0Font can be descendant from any font, we need to remember the original TrueTypeFont for the glyph information
-            this.type0Font = PDType0Font.load(document, this.trueTypeFont, false); // use Type0Font for unicode support
+        try (var in = Thread.currentThread().getContextClassLoader().getResourceAsStream(resourcePath); var buffer = new RandomAccessReadBuffer(in)) {
+            TrueTypeFont trueTypeFont = new TTFParser().parse(buffer); // since Type0Font can be descendant from any font, we need to remember the original TrueTypeFont for the glyph information
+            PDType0Font type0Font = PDType0Font.load(document, trueTypeFont, true); // use Type0Font for unicode support
+            return new Type0FontMetricsFactory(type0Font, trueTypeFont);
        }
    }

@@ -55,8 +77,9 @@ public class Type0FontMetricsFactory implements FontMetricsFactory {
                if (glyph == null || glyph.getBoundingBox() == null) {
                    continue;
                }
-
-                descent = Math.min(descent, glyph.getYMinimum());
+                if (!slashGlyphIds.contains(glyphId)) {
+                    descent = Math.min(descent, glyph.getYMinimum());
+                }
                height = Math.max(height, glyph.getYMaximum());
            } catch (Exception e) {
                log.warn("descent and height of string {} could not be parsed, using average fallback value!", text);
@@ -0,0 +1,158 @@
+package com.knecon.fforesight.service.ocr.processor.service.scriptdetection;
+
+import java.util.Comparator;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Optional;
+import java.util.stream.Stream;
+
+import org.apache.commons.math3.ml.clustering.Cluster;
+import org.apache.commons.math3.ml.clustering.DBSCANClusterer;
+import org.apache.pdfbox.pdmodel.PDDocument;
+import org.springframework.stereotype.Service;
+
+import com.knecon.fforesight.service.ocr.processor.model.OcrResult;
+import com.knecon.fforesight.service.ocr.processor.model.OcrResultToWrite;
+import com.knecon.fforesight.service.ocr.processor.model.scriptdetection.FontStyleDetectionModel;
+import com.knecon.fforesight.service.ocr.processor.model.scriptdetection.TextPositionAndWordImage;
+import com.knecon.fforesight.service.ocr.processor.model.scriptdetection.WordImage;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.FontMetricsFactory;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.FontStyle;
+import com.knecon.fforesight.service.ocr.processor.service.fonts.Type0FontMetricsFactory;
+import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;
+
+import lombok.AccessLevel;
+import lombok.RequiredArgsConstructor;
+import lombok.experimental.FieldDefaults;
+import lombok.extern.slf4j.Slf4j;
+
+@Slf4j
+@Service
+@RequiredArgsConstructor
+@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
+public class FontStyleDetector {
+
+    OcrServiceSettings settings;
+    StrokeWidthCalculator strokeWidthCalculator;
+
+
+    /**
+     * Implementation of the MOBDoB algorithm, refer to the paper here:
+     * <a href="http://mile.ee.iisc.ac.in/publications/softCopy/DocumentAnalysis/Sai_NCVPRIPG2013.pdf">Script Independent Detection of Bold Words in Multi Font-size Documents</a>
+     * <p>
+     * As a high level overview: We cluster all text based on its font size. We determine the cluster with the most words. This is assumed to be regular text.
+     * We then estimate the average stroke width of that cluster by thinning all text to a single pixel and calculating the ratio of remaining pixels.
+     * (<a href="http://www.leptonica.org/papers/conn.pdf">Leptonica Documentation on thinning</a>)
+     * For each word we scale this average strokewidth based on its fontsize compared to the most common fontsize.
+     * Using the scaled strokewidth we do an opening operation.
+     * (<a href="https://en.wikipedia.org/wiki/Opening_(morphology)">Opening (Morphology)</a>).
+     * We then threshold the ratio of remaining pixels to determine whether a word is bold or not.
+     * <p>
+     * I did take some liberties though. Firstly, the paper uses text height without ascender/descender height for the clustering. I'm using the previously implemented font size.
+     * But this is based on text width. Thus, I'm also using the height scaling factor to scale the font size by the text height.
+     * The paper does not describe its clustering algorithm, so I've decided on DBSCAN due to its good runtime and readily available implementation by apache commons math.
+     * Moreover, the paper states that stroke width scales linearly with text height. I've come to the conclusion this is not the case.
+     * It seems it scales with the square root of the text height. Or at least this seemed to give the best results.
+     */
+    public Map<Integer, List<OcrResultToWrite>> detectBold(List<OcrResult> ocrResults, PDDocument document) {
+
+        FontMetricsFactory fontMetricsFactory = Type0FontMetricsFactory.regular(document);
+        if (!settings.isBoldDetection()) {
+            return OcrResultToWrite.buildOcrResultsToWrite(ocrResults, fontMetricsFactory);
+        }
+
+        Map<Integer, List<OcrResultToWrite>> ocrResultToWritePerPage = new HashMap<>();
+
+        DBSCANClusterer<TextPositionAndWordImage> clusterer = new DBSCANClusterer<>(0.5, 1);
+
+        FontMetricsFactory boldFontMetricsFactory = Type0FontMetricsFactory.bold(document);
+
+        for (OcrResult result : ocrResults) {
+            FontStyleDetectionModel fontStyleDetectionModel = FontStyleDetectionModel.fromOcrResult(result, fontMetricsFactory, settings);
+
+            List<Cluster<TextPositionAndWordImage>> clusters = clusterer.cluster(fontStyleDetectionModel.getTextPositionsAndWordImages());
+            Optional<Cluster<TextPositionAndWordImage>> largestCluster = clusters.stream().max(Comparator.comparingInt(cluster -> cluster.getPoints().size()));
+
+            if (largestCluster.isEmpty()) {
+                insertResultIntoMap(result.image().getPageNumber(), ocrResultToWritePerPage, fontStyleDetectionModel);
+                continue;
+            }
+
+            List<TextPositionAndWordImage> wordsWithMostCommonTextHeight = largestCluster.get().getPoints();
+
+            double standardTextHeight = calculateStandardTextheight(wordsWithMostCommonTextHeight);
+            double regularStrokeWidth = calculateRegularStrokeWidth(wordsWithMostCommonTextHeight);
+
+            for (TextPositionAndWordImage textPositionsAndWordImage : fontStyleDetectionModel.getTextPositionsAndWordImages()) {
+                decideOnFontStyle(textPositionsAndWordImage, regularStrokeWidth, standardTextHeight, boldFontMetricsFactory);
+            }
+
+            insertResultIntoMap(result.image().getPageNumber(), ocrResultToWritePerPage, fontStyleDetectionModel);
+            fontStyleDetectionModel.dispose();
+        }
+
+        log.info("Finished bold detection");
+        return ocrResultToWritePerPage;
+    }
+
+
+    private static double calculateStandardTextheight(List<TextPositionAndWordImage> wordsWithMostCommonTextHeight) {
+
+        return wordsWithMostCommonTextHeight.stream()
+                .map(TextPositionAndWordImage::getWordImage)
+                .mapToDouble(WordImage::getTextHeight)
+                .filter(Double::isFinite)
+                .average()
+                .orElseThrow();
+    }
+
+
+    private double calculateRegularStrokeWidth(List<TextPositionAndWordImage> wordsWithMostCommonTextHeight) {
+
+        return wordsWithMostCommonTextHeight.stream()
+                .mapToDouble(textPositionAndWordImage -> strokeWidthCalculator.calculate(textPositionAndWordImage.getWordImage().getImage()))
+                .filter(Double::isFinite)
+                .average()
+                .orElseThrow();
+    }
+
+
+    private static void insertResultIntoMap(int pageNumber, Map<Integer, List<OcrResultToWrite>> ocrResultToWritePerPage, FontStyleDetectionModel fontStyleDetectionModel) {
+
+        OcrResultToWrite ocrResult = OcrResultToWrite.fromFontStyleDetectionModel(fontStyleDetectionModel);
+
+        ocrResultToWritePerPage.compute(pageNumber, (key, existingList) -> {
+            if (existingList == null) {
+                return List.of(ocrResult);
+            } else {
+                return Stream.concat(existingList.stream(), Stream.of(ocrResult)).toList();
+            }
+        });
+    }
+
+
+    private void decideOnFontStyle(TextPositionAndWordImage textPositionsAndWordImage,
+                                   double standardStrokeWidth,
+                                   double standardTextHeight,
+                                   FontMetricsFactory boldFontMetricsFactory) {
+
+        double scaledStrokeWidth = scaleStrokeWidthByFontSize(textPositionsAndWordImage, standardStrokeWidth, standardTextHeight);
+
+        if (textPositionsAndWordImage.getWordImage().hasLargerStrokeWidth(scaledStrokeWidth)) {
+            textPositionsAndWordImage.getTextPositionInImage().setFontMetricsFactory(boldFontMetricsFactory);
+            textPositionsAndWordImage.getTextPositionInImage().setFontStyle(FontStyle.BOLD);
+        } else {
+            textPositionsAndWordImage.getTextPositionInImage().setFontStyle(FontStyle.REGULAR);
+        }
+    }
+
+
+    private static double scaleStrokeWidthByFontSize(TextPositionAndWordImage textPositionsAndWordImage, double standardStrokeWidth, double standardFontSize) {
+
+        double influenceOfFontSize = 1.0; // the paper states that stroke width scales exactly linearly with font size. This did not seem to be true for me. Maybe some of the preprocessing steps are affecting this.
+        double fontsizeScalingFactor = Math.sqrt(textPositionsAndWordImage.getWordImage().getTextHeight() / standardFontSize);
+        return standardStrokeWidth + (influenceOfFontSize * (fontsizeScalingFactor - 1) * standardStrokeWidth);
+    }
+
+}
@@ -0,0 +1,57 @@
+package com.knecon.fforesight.service.ocr.processor.service.scriptdetection;
+
+import com.knecon.fforesight.service.ocr.processor.utils.ImageProcessingUtils;
+
+import lombok.AccessLevel;
+import lombok.NoArgsConstructor;
+import lombok.experimental.FieldDefaults;
+import net.sourceforge.lept4j.Leptonica1;
+import net.sourceforge.lept4j.Pix;
+import net.sourceforge.lept4j.Sel;
+import net.sourceforge.lept4j.util.LeptUtils;
+
+/**
+ * This code is a good start for detecting italic text, although it has a few issues especially with glyphs which are naturally slanted. E.g. z, 2, 7, /
+ * If we want this maybe we should exclude these glyphs and then it might have less false positives. But in its current state i don't recommend using it.
+ */
+@NoArgsConstructor
+@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
+public class ItalicDetector {
+
+
+    static String italicKernel = "ooxxooxxooxxoxxooXxooxxoxxooxxooxxoo";
+    Sel italicSel = Leptonica1.selCreateFromString(italicKernel, 9, 4, "italicKernel");
+    Sel brickSel = Leptonica1.selCreateBrick(3, 4, 1, 2, 1);
+
+
+    public boolean isItalic(Pix pix) {
+
+        Pix preprocessed = preprocess(pix);
+        Pix flipped = Leptonica1.pixFlipLR(null, pix);
+        Pix flippedPreprocessed = preprocess(flipped);
+        Leptonica1.pixFlipLR(flippedPreprocessed, flippedPreprocessed);
+        double pixelDensity = ImageProcessingUtils.calculatePixelDensity(preprocessed);
+        double flippedPixelDensity = ImageProcessingUtils.calculatePixelDensity(flippedPreprocessed);
+        LeptUtils.disposePix(preprocessed);
+        LeptUtils.disposePix(flipped);
+        LeptUtils.disposePix(flippedPreprocessed);
+        return flippedPixelDensity / pixelDensity < 0.85;
+    }
+
+
+    private Pix preprocess(Pix pix) {
+
+        Pix eroded = Leptonica1.pixErode(null, pix, italicSel.getPointer());
+        Pix dilated = Leptonica1.pixDilate(null, eroded, brickSel.getPointer());
+        LeptUtils.disposePix(eroded);
+        return dilated;
+    }
+
+
+    public void dispose() {
+
+        LeptUtils.dispose(italicSel);
+        LeptUtils.dispose(brickSel);
+    }
+
+}
@@ -0,0 +1,58 @@
+package com.knecon.fforesight.service.ocr.processor.service.scriptdetection;
+
+import static net.sourceforge.lept4j.ILeptonica.L_THIN_FG;
+
+import java.nio.IntBuffer;
+
+import org.springframework.stereotype.Service;
+
+import lombok.AccessLevel;
+import lombok.NoArgsConstructor;
+import lombok.experimental.FieldDefaults;
+import net.sourceforge.lept4j.Leptonica1;
+import net.sourceforge.lept4j.Pix;
+import net.sourceforge.lept4j.Sela;
+import net.sourceforge.lept4j.util.LeptUtils;
+
+@Service
+@NoArgsConstructor
+@FieldDefaults(level = AccessLevel.PRIVATE)
+public class StrokeWidthCalculator {
+
+    Sela thinningSel;
+
+
+    /**
+     * Uses a series of sels to thin all connected lines to a single pixel. Then the pixel ratio is a good estimation of the stroke width in pixels.
+     * <a href="http://www.leptonica.org/papers/conn.pdf">Leptonica Documentation on thinning</a>
+     * Since the baseline is a strokewidth of exactly one, we need to add 1 to the result.
+     *
+     * @param input binarized pix with text on it
+     * @return estimated stroke width in pixels
+     */
+    public double calculate(Pix input) {
+
+        init();
+
+        Pix thinned = Leptonica1.pixThinConnectedBySet(input, L_THIN_FG, thinningSel, 0);
+
+        IntBuffer thinnedPixelCount = IntBuffer.allocate(1);
+        Leptonica1.pixCountPixels(thinned, thinnedPixelCount, null);
+
+        IntBuffer pixelCount = IntBuffer.allocate(1);
+        Leptonica1.pixCountPixels(input, pixelCount, null);
+
+        LeptUtils.disposePix(thinned);
+
+        return (double) pixelCount.get() / thinnedPixelCount.get() + 1;
+    }
+
+
+    private void init() {
+
+        if (thinningSel == null) {
+            thinningSel = Leptonica1.selaMakeThinSets(1, 0);
+        }
+    }
+
+}
@@ -0,0 +1,65 @@
+package com.knecon.fforesight.service.ocr.processor.service.threads;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.NoSuchElementException;
+import java.util.concurrent.BlockingQueue;
+
+import com.knecon.fforesight.service.ocr.processor.model.OcrImage;
+import com.knecon.fforesight.service.ocr.processor.model.RenderedPageImageFile;
+import com.knecon.fforesight.service.ocr.processor.model.UnprocessedImage;
+
+import lombok.AccessLevel;
+import lombok.RequiredArgsConstructor;
+import lombok.Setter;
+import lombok.SneakyThrows;
+import lombok.experimental.FieldDefaults;
+import lombok.extern.slf4j.Slf4j;
+import net.sourceforge.tess4j.TessAPI1;
+
+/*
+This just moves the Elements from the GhostScriptOutputListener into the ImageProcessing queue asynchronously
+ */
+@Slf4j
+@RequiredArgsConstructor
+@FieldDefaults(level = AccessLevel.PRIVATE)
+public class BlockingQueueFiller extends Thread {
+
+    final BlockingQueue<RenderedPageImageFile> imageInputQueue;
+    final BlockingQueue<UnprocessedImage> imageOutputQueue;
+
+    @Setter
+    boolean allImagesQueued;
+
+
+    @SneakyThrows
+    @Override
+    public void run() {
+
+        // Interrupting signals that the image extraction has finished
+        try {
+            while (!allImagesQueued) {
+                final UnprocessedImage image = imageInputQueue.take();
+                try {
+                    imageOutputQueue.put(image);
+                } catch (InterruptedException e) {
+                    imageOutputQueue.put(image);
+                }
+            }
+        } catch (InterruptedException e) {
+            log.info("All images extracted, emptying processing queue and stopping");
+        }
+
+        // empty the queue
+        try {
+            while (true) {
+                final UnprocessedImage image = imageInputQueue.remove();
+                imageOutputQueue.put(image);
+            }
+        } catch (NoSuchElementException e) {
+            log.debug("No images left in queue, stopping.");
+        }
+
+    }
+
+}
@@ -0,0 +1,122 @@
+package com.knecon.fforesight.service.ocr.processor.service.threads;
+
+import java.io.BufferedReader;
+import java.io.File;
+import java.io.InputStream;
+import java.io.InputStreamReader;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Queue;
+import java.util.concurrent.BlockingQueue;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
+
+import com.knecon.fforesight.service.ocr.processor.model.RenderedPageImageFile;
+
+import lombok.AccessLevel;
+import lombok.RequiredArgsConstructor;
+import lombok.SneakyThrows;
+import lombok.experimental.FieldDefaults;
+import lombok.extern.slf4j.Slf4j;
+
+@Slf4j
+@RequiredArgsConstructor
+@FieldDefaults(level = AccessLevel.PRIVATE)
+public class GhostScriptOutputHandler extends Thread {
+
+    static Pattern pageFinishedPattern = Pattern.compile("Page (\\d+)");
+
+    // If the stdError or stdOut buffer of a thread is not being emptied it might lock the process in case of errors, so we need to empty both streams to prevent a deadlock.
+    // Since both need to read simultaneously we need to implement the readers as separate threads.
+
+    final InputStream is;
+    final String processName;
+    final Type type;
+
+    final Map<Integer, RenderedPageImageFile> pagesToProcess;
+    final BlockingQueue<RenderedPageImageFile> renderedPageImageFileOutput;
+
+    int currentPageNumber;
+
+
+    public static GhostScriptOutputHandler errorHandler(InputStream is) {
+
+        return new GhostScriptOutputHandler(is, "GS", Type.ERROR, null, null);
+    }
+
+
+    public static GhostScriptOutputHandler stdOut(InputStream is,
+                                                  Map<Integer, RenderedPageImageFile> pagesToProcess,
+                                                  BlockingQueue<RenderedPageImageFile> renderedPageImageFileOutput) {
+
+        return new GhostScriptOutputHandler(is, "GS", Type.STD_OUT, pagesToProcess, renderedPageImageFileOutput);
+    }
+
+
+    @SneakyThrows
+    public void run() {
+
+        try (InputStreamReader isr = new InputStreamReader(is); BufferedReader br = new BufferedReader(isr)) {
+
+            String line;
+            while (true) {
+                line = br.readLine();
+
+                if (line == null) {
+                    break;
+                }
+
+                if (type.equals(Type.ERROR)) {
+                    log.error(processName + "_" + type.name() + ">" + line);
+                } else {
+                    log.debug(processName + "_" + type.name() + ">" + line);
+                    addProcessedImageToQueue(line);
+                }
+            }
+        }
+        is.close();
+        if (type.equals(Type.STD_OUT)) {
+            queueFinishedPage(currentPageNumber);
+        }
+
+    }
+
+
+    private void addProcessedImageToQueue(String line) {
+
+        /*
+        Ghostscript prints the pageNumber it is currently working on, so we remember the current page and queue it as soon as the next comes in.
+         */
+        Matcher pageNumberMatcher = pageFinishedPattern.matcher(line);
+        if (pageNumberMatcher.find()) {
+            int pageNumber = Integer.parseInt(pageNumberMatcher.group(1));
+
+            if (currentPageNumber == 0) {
+                currentPageNumber = pageNumber;
+                return;
+            }
+
+            queueFinishedPage(currentPageNumber);
+            currentPageNumber = pageNumber;
+        }
+    }
+
+
+    private void queueFinishedPage(int pageNumber) {
+
+        var imageFile = this.pagesToProcess.get(pageNumber);
+        if (imageFile == null) {
+            throw new IllegalArgumentException(String.format("Page number %d does not exist in this thread. It only has pagenumbers %s", pageNumber, pagesToProcess.keySet()));
+        }
+        assert new File(imageFile.absoluteFilePath()).isFile();
+        renderedPageImageFileOutput.add(imageFile);
+    }
+
+
+    public enum Type {
+        ERROR,
+        STD_OUT
+    }
+
+}
@@ -5,12 +5,11 @@ import java.util.List;
 import java.util.concurrent.BlockingQueue;

 import org.apache.pdfbox.Loader;
-import org.apache.pdfbox.io.MemoryUsageSetting;
 import org.apache.pdfbox.pdmodel.PDDocument;
 import org.apache.pdfbox.pdmodel.PDPage;

-import com.knecon.fforesight.service.ocr.processor.model.ExtractedOcrImage;
-import com.knecon.fforesight.service.ocr.processor.model.OcrImage;
+import com.knecon.fforesight.service.ocr.processor.model.ExtractedImage;
+import com.knecon.fforesight.service.ocr.processor.model.UnprocessedImage;
 import com.knecon.fforesight.service.ocr.processor.service.ImageStreamEngine;
 import com.knecon.fforesight.service.ocr.processor.service.OcrProgressLogger;
 import com.knecon.fforesight.service.ocr.processor.service.Statistics;
@@ -26,6 +25,7 @@ import lombok.experimental.FieldDefaults;
@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
 public class ImageExtractionThread extends Thread {

+    static double FULL_PAGE_IMAGE_THRESHOLD = 0.99;
    static double IMAGE_ALIGNMENT_THRESHOLD = 1;

    int id;
@@ -37,9 +37,10 @@ public class ImageExtractionThread extends Thread {
    OcrServiceSettings settings;

    // output is written to these lists
-    BlockingQueue<OcrImage> imageOutputQueue;
+    BlockingQueue<UnprocessedImage> imageProcessingQueue;
    List<Integer> stitchedPageNumbers;

+
    @SneakyThrows
    @Override
    public void run() {
@@ -48,28 +49,28 @@ public class ImageExtractionThread extends Thread {
        for (Integer pageIndex : pageIndices) {
            try (PDDocument document = Loader.loadPDF(documentFile)) { // load new PDDocument for thread safety, also keeps RAM usage low.
                timestamp = System.currentTimeMillis();
-                List<ExtractedOcrImage> extractedOcrImages = getExtractedOcrImages(pageIndex, document);
+                List<ExtractedImage> extractedImages = getExtractedImages(pageIndex, document);
                stats.increaseImageExtraction(id, System.currentTimeMillis() - timestamp);
-                if (extractedOcrImages.isEmpty()) {
+                if (extractedImages.isEmpty()) {
                    logger.logPageSkipped(pageIndex);
                }

-                if (checkForStitchedImages(extractedOcrImages)) {
+                if (checkForFullPageOrStitchedImages(extractedImages, document.getPage(pageIndex - 1))) {
                    stitchedPageNumbers.add(pageIndex);
                    logger.addImagesToProcess(pageIndex, 0);
                    continue;
                }

-                for (ExtractedOcrImage image : extractedOcrImages) {
-                    imageOutputQueue.put(image);
-                    logger.addImagesToProcess(image.getPageNumber(), image.getNumberOnPage());
+                for (ExtractedImage image : extractedImages) {
+                    imageProcessingQueue.put(image);
+                    logger.addImagesToProcess(image.pageNumber(), image.numberOnPage());
                }
            }
        }
    }


-    private List<ExtractedOcrImage> getExtractedOcrImages(Integer pageIndex, PDDocument document) {
+    private List<ExtractedImage> getExtractedImages(Integer pageIndex, PDDocument document) {

        PDPage page = document.getPage(pageIndex - 1);
        ImageStreamEngine imageStreamEngine = new ImageStreamEngine(settings);
@@ -79,22 +80,25 @@ public class ImageExtractionThread extends Thread {


    @SneakyThrows
-    private boolean checkForStitchedImages(List<ExtractedOcrImage> imagesOnCurrentPage) {
+    private boolean checkForFullPageOrStitchedImages(List<ExtractedImage> imagesOnCurrentPage, PDPage page) {

-        if (imagesOnCurrentPage.size() <= 1) {
+        if (imagesOnCurrentPage.isEmpty()) {
            return false;
        }

-        //checking for intersections or direct alignment of images
-        ExtractedOcrImage[] imageOnPagesArray = new ExtractedOcrImage[imagesOnCurrentPage.size()];
-        int index = 0;
-        for (ExtractedOcrImage imageOnPage : imagesOnCurrentPage) {
-            imageOnPagesArray[index] = imageOnPage;
-            index++;
+        for (ExtractedImage imageOnPage : imagesOnCurrentPage) {
+            if (imageOnPage.width() > FULL_PAGE_IMAGE_THRESHOLD * page.getCropBox().getWidth() && imageOnPage.height() > FULL_PAGE_IMAGE_THRESHOLD * page.getCropBox()
+                    .getHeight()) {
+                return true;
+            }
        }
-        for (int j = 0; j < imageOnPagesArray.length; j++) {
-            for (int i = j + 1; i < imageOnPagesArray.length; i++) {
-                if (imageOnPagesArray[j].getImageCoordinatesInInitialUserSpace().aligns(imageOnPagesArray[i].getImageCoordinatesInInitialUserSpace(), IMAGE_ALIGNMENT_THRESHOLD)) {
+
+        //checking for intersections or direct alignment of images
+        for (int j = 0; j < imagesOnCurrentPage.size(); j++) {
+            for (int i = j + 1; i < imagesOnCurrentPage.size(); i++) {
+                if (imagesOnCurrentPage.get(j)
+                        .getImageCoordinatesInInitialUserSpace()
+                        .aligns(imagesOnCurrentPage.get(i).getImageCoordinatesInInitialUserSpace(), IMAGE_ALIGNMENT_THRESHOLD)) {
                    // TODO: see if we can stitch aligning images using BufferedImage and skip the gs conversion entirely
                    return true;
                }
@@ -0,0 +1,250 @@
+package com.knecon.fforesight.service.ocr.processor.service.threads;
+
+import static net.sourceforge.tess4j.ITessAPI.TRUE;
+
+import java.nio.FloatBuffer;
+import java.nio.IntBuffer;
+import java.util.NoSuchElementException;
+import java.util.concurrent.BlockingQueue;
+
+import org.apache.pdfbox.pdmodel.PDDocument;
+
+import com.knecon.fforesight.service.ocr.processor.model.ExtractedImage;
+import com.knecon.fforesight.service.ocr.processor.model.ExtractedOcrImage;
+import com.knecon.fforesight.service.ocr.processor.model.OcrImage;
+import com.knecon.fforesight.service.ocr.processor.model.PageInformation;
+import com.knecon.fforesight.service.ocr.processor.model.RenderedPageImageFile;
+import com.knecon.fforesight.service.ocr.processor.model.RenderedPageOcrImage;
+import com.knecon.fforesight.service.ocr.processor.model.UnprocessedImage;
+import com.knecon.fforesight.service.ocr.processor.service.Statistics;
+import com.knecon.fforesight.service.ocr.processor.settings.OcrServiceSettings;
+import com.knecon.fforesight.service.ocr.processor.utils.ImageProcessingUtils;
+import com.sun.jna.ptr.PointerByReference;
+
+import lombok.AccessLevel;
+import lombok.RequiredArgsConstructor;
+import lombok.Setter;
+import lombok.SneakyThrows;
+import lombok.experimental.FieldDefaults;
+import lombok.extern.slf4j.Slf4j;
+import net.sourceforge.lept4j.L_Kernel;
+import net.sourceforge.lept4j.Leptonica1;
+import net.sourceforge.lept4j.Pix;
+import net.sourceforge.lept4j.util.LeptUtils;
+import net.sourceforge.tess4j.ITessAPI;
+import net.sourceforge.tess4j.TessAPI1;
+
+/*
+ * This thread does all the image processing. There should only be one, since Leptonica is not thread safe.
+ */
+@Slf4j
+@RequiredArgsConstructor
+@FieldDefaults(level = AccessLevel.PRIVATE)
+public class ImageProcessingThread extends Thread {
+
+    final BlockingQueue<UnprocessedImage> imageInputQueue;
+    final BlockingQueue<OcrImage> imageOutputQueue;
+    final ITessAPI.TessBaseAPI detectionScriptHandle = initDetectionScriptHandle();
+    final L_Kernel gaussianKernel = Leptonica1.makeGaussianKernel(2, 2, 1.2f, 1);
+    final Statistics stats;
+    final OcrServiceSettings settings;
+    final PDDocument document;
+
+    @Setter
+    boolean allImagesExtracted;
+
+
+    @SneakyThrows
+    @Override
+    public void run() {
+
+        try {
+            while (!allImagesExtracted) {
+                final UnprocessedImage image = imageInputQueue.take();
+                var ocrImage = this.process(image);
+                try {
+                    imageOutputQueue.put(ocrImage);
+                } catch (InterruptedException e) {
+                    imageOutputQueue.put(ocrImage);
+                }
+            }
+        } catch (InterruptedException e) {
+            log.info("All images extracted, emptying processing queue and stopping");
+        }
+
+        try {
+            while (true) {
+                final UnprocessedImage image = imageInputQueue.remove();
+                OcrImage ocrImage = this.process(image);
+                imageOutputQueue.put(ocrImage);
+            }
+        } catch (NoSuchElementException e) {
+            log.debug("No images left in processing queue, stopping.");
+        }
+
+        TessAPI1.TessBaseAPIEnd(this.detectionScriptHandle);
+        TessAPI1.TessBaseAPIDelete(this.detectionScriptHandle);
+        LeptUtils.dispose(gaussianKernel);
+    }
+
+
+    private OcrImage process(UnprocessedImage unprocessedImage) {
+
+        long timestamp = System.currentTimeMillis();
+
+        OcrImage ocrImage;
+        if (unprocessedImage instanceof ExtractedImage extractedImage) {
+            ocrImage = processExtractedImage(extractedImage);
+        } else if (unprocessedImage instanceof RenderedPageImageFile renderedPageImageFile) {
+            ocrImage = processRenderedPageImageFile(renderedPageImageFile);
+        } else {
+            throw new UnsupportedOperationException(String.format("Class %s is not supported!", unprocessedImage.getClass()));
+        }
+
+        stats.increaseImageProcessing(System.currentTimeMillis() - timestamp);
+
+        return ocrImage;
+    }
+
+
+    private OcrImage processRenderedPageImageFile(RenderedPageImageFile renderedPageImageFile) {
+
+        Pix pix = processPix(renderedPageImageFile.asPix(), settings.getDpi(), settings.getDpi());
+
+        int orientDegree = detectOrientation(pix, settings.getDpi(), detectionScriptHandle);
+        Pix rotatedPix = ImageProcessingUtils.deRotatePix(orientDegree, pix);
+
+        OcrImage ocrImage = new RenderedPageOcrImage(pix.h,
+                pix.w,
+                PageInformation.fromPDPage(renderedPageImageFile.pageNumber(), document.getPage(renderedPageImageFile.pageNumber() - 1)),
+                rotatedPix,
+                orientDegree);
+
+        if (pix != rotatedPix) {
+            LeptUtils.disposePix(pix);
+        }
+
+        return ocrImage;
+    }
+
+
+    private OcrImage processExtractedImage(ExtractedImage extractedImage) {
+
+        float imageDPI = Math.abs(extractedImage.image().getWidth() / (extractedImage.ctm().getScalingFactorX() / 72));
+
+        Pix pix = processPix(extractedImage.asPix(), imageDPI, settings.getDpi());
+
+        int orientDegree = detectOrientation(pix, settings.getDpi(), detectionScriptHandle);
+        Pix rotatedPix = ImageProcessingUtils.deRotatePix(orientDegree, pix);
+
+        OcrImage ocrImage = new ExtractedOcrImage(extractedImage.pageNumber(),
+                extractedImage.numberOnPage(),
+                extractedImage.height(),
+                extractedImage.width(),
+                extractedImage.ctm(),
+                rotatedPix,
+                pix.h,
+                pix.w,
+                orientDegree);
+
+        if (pix != rotatedPix) {
+            LeptUtils.disposePix(pix);
+        }
+        return ocrImage;
+    }
+
+
+    public int detectOrientation(Pix pix, int dpi, ITessAPI.TessBaseAPI detectionScriptHandle) {
+
+        TessAPI1.TessBaseAPISetImage2(detectionScriptHandle, pix);
+        TessAPI1.TessBaseAPISetSourceResolution(detectionScriptHandle, dpi);
+
+        IntBuffer orientationDegreeResultBuffer;
+        FloatBuffer orientationDegreeConfidenceBuffer;
+        PointerByReference scriptureNameBuffer;
+        FloatBuffer scriptureConfidenceBuffer;
+
+        orientationDegreeResultBuffer = IntBuffer.allocate(1);
+        orientationDegreeConfidenceBuffer = FloatBuffer.allocate(1);
+        scriptureNameBuffer = new PointerByReference(); // Is this memory being freed?
+        scriptureConfidenceBuffer = FloatBuffer.allocate(1);
+
+        int orientationDegree = 0;
+        int result = TessAPI1.TessBaseAPIDetectOrientationScript(detectionScriptHandle,
+                orientationDegreeResultBuffer,
+                orientationDegreeConfidenceBuffer,
+                scriptureNameBuffer,
+                scriptureConfidenceBuffer);
+        if (result == TRUE && orientationDegreeConfidenceBuffer.get() > settings.getMinRotationConfidence()) {
+            orientationDegree = orientationDegreeResultBuffer.get();
+        }
+
+        TessAPI1.TessBaseAPIClear(detectionScriptHandle);
+
+        return orientationDegree;
+    }
+
+
+    @SneakyThrows
+    private Pix processPix(Pix pix, float imageDpi, int targetDpi) {
+
+        Pix grayScale;
+        Pix scaledUp;
+        Pix gaussian;
+        Pix binarized;
+
+        //convert to grayscale
+        if (pix.d == 8) {
+            grayScale = pix;
+        } else if (pix.d == 32) {
+            grayScale = Leptonica1.pixConvertRGBToGrayFast(pix);
+        } else if (pix.d == 1) {
+            grayScale = Leptonica1.pixConvert1To8(null, pix, (byte) 0, (byte) 255);
+        } else {
+            throw new UnsupportedOperationException(String.format("Unknown pix format with bpp of %d", pix.d));
+        }
+
+        // scale up
+        float targetFactor = targetDpi / imageDpi;
+        if (targetFactor > 2.1) {
+            scaledUp = Leptonica1.pixScaleGray4xLI(grayScale);
+        } else if (targetFactor > 1.1) {
+            scaledUp = Leptonica1.pixScaleGray2xLI(grayScale);
+        } else {
+            scaledUp = grayScale;
+        }
+
+        // remove noise and prep for Otsu
+        gaussian = Leptonica1.pixConvolve(scaledUp, gaussianKernel, 8, 1);
+
+        // Threshold to binary
+        if (pix.w < 100 || pix.h < 100) {
+            binarized = Leptonica1.pixThresholdToBinary(gaussian, 170);
+        } else {
+            binarized = Leptonica1.pixOtsuThreshOnBackgroundNorm(gaussian, null, 50, 50, 165, 10, 100, 5, 5, 0.2f, null);
+
+            if (binarized == null) { // Sometimes Otsu just fails, then we binarize directly
+                binarized = Leptonica1.pixThresholdToBinary(gaussian, 170);
+            }
+        }
+
+        LeptUtils.disposePix(pix);
+        LeptUtils.disposePix(grayScale);
+        LeptUtils.disposePix(scaledUp);
+        LeptUtils.disposePix(gaussian);
+
+        return binarized;
+    }
+
+
+
+    private static ITessAPI.TessBaseAPI initDetectionScriptHandle() {
+
+        ITessAPI.TessBaseAPI handle = TessAPI1.TessBaseAPICreate();
+        String datapath = System.getenv("TESSDATA_PREFIX");
+        TessAPI1.TessBaseAPIInit3(handle, datapath, "osd");
+        TessAPI1.TessBaseAPISetVariable(handle, "debug_file", "/dev/null");
+        return handle;
+    }
+
+}
@@ -1,6 +1,10 @@
 package com.knecon.fforesight.service.ocr.processor.service.threads;

 import static net.sourceforge.tess4j.ITessAPI.TRUE;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPICreate;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPIInit1;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPISetPageSegMode;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPISetVariable;

 import java.io.File;
 import java.nio.FloatBuffer;
@@ -16,6 +20,7 @@ import com.knecon.fforesight.service.ocr.processor.model.OcrResult;
 import com.knecon.fforesight.service.ocr.processor.service.OcrProgressLogger;
 import com.knecon.fforesight.service.ocr.processor.service.Statistics;
 import com.knecon.fforesight.service.ocr.processor.utils.Tesseract2;
+import com.sun.jna.StringArray;
 import com.sun.jna.ptr.PointerByReference;

 import lombok.AccessLevel;
@@ -43,7 +48,6 @@ public class OCRThread extends Thread {
    Statistics stats;
    OcrServiceSettings settings;
    Tesseract2 instance;
-    ITessAPI.TessBaseAPI detectionScriptHandle;


    public OCRThread(int id,
@@ -62,7 +66,6 @@ public class OCRThread extends Thread {
        this.stats = stats;
        this.settings = settings;
        this.instance = createInstance(settings);
-        this.detectionScriptHandle = initDetectionScriptHandle();
    }


@@ -87,8 +90,9 @@ public class OCRThread extends Thread {
                this.process(image);
            }
        } catch (NoSuchElementException e) {
-            log.debug("Processed all Images, finishing.");
+            log.debug("Executed tesseract on all Images, finishing.");
        }
+
    }


@@ -100,12 +104,8 @@ public class OCRThread extends Thread {

        int psm = settings.getPsmOverride() < 0 ? image.getOptimalPageSegmentationMode() : settings.getPsmOverride();

-        int orientDegree = detectOrientation(image);
-        image.setRotationDegrees(orientDegree);
-        Pix rotatedPix = image.getRotatedPix();
-        executeTesseract(psm, image.getDpi(), rotatedPix, tesseractOutputFileName);
+        executeTesseract(psm, image.getDpi(), image.getPix(), tesseractOutputFileName);
        image.destroyPix();
-        LeptUtils.disposePix(rotatedPix);

        results.add(OcrResult.create(image, tesseractOutputFileName));
        logger.logImageFinished(image, psm);
@@ -113,50 +113,14 @@ public class OCRThread extends Thread {
    }


-    public int detectOrientation(OcrImage image) {
-
-        TessAPI1.TessBaseAPISetImage2(detectionScriptHandle, image.getPix());
-        TessAPI1.TessBaseAPISetSourceResolution(detectionScriptHandle, image.getDpi());
-
-        IntBuffer orient_degB = IntBuffer.allocate(1);
-        FloatBuffer orient_confB = FloatBuffer.allocate(1);
-        PointerByReference script_nameB = new PointerByReference();
-        FloatBuffer script_confB = FloatBuffer.allocate(1);
-
-        int orient_deg = 0;
-        int result = TessAPI1.TessBaseAPIDetectOrientationScript(detectionScriptHandle, orient_degB, orient_confB, script_nameB, script_confB);
-        if (result == TRUE) {
-            orient_deg = orient_degB.get();
-        }
-        TessAPI1.TessBaseAPIClear(detectionScriptHandle);
-
-        return orient_deg;
-    }
-
-
-    private static ITessAPI.TessBaseAPI initDetectionScriptHandle() {
-
-        ITessAPI.TessBaseAPI handle = TessAPI1.TessBaseAPICreate();
-        String datapath = System.getenv("TESSDATA_PREFIX");
-        TessAPI1.TessBaseAPIInit3(handle, datapath, "osd");
-
-        return handle;
-    }
-
-
    @SneakyThrows
    public void executeTesseract(int psm, int dpi, Pix pix, String tesseractOutputFileName) {

-        if (settings.isDebug()) {
-            String[] a = tesseractOutputFileName.split("/");
-            String folder = "/tmp/pixs/" + a[a.length - 3];
-            new File(folder).mkdirs();
-            Leptonica1.pixWrite(folder + "/pix_" + a[a.length - 1] + ".png", pix, 3);
-        }
-
+        Leptonica1.pixWrite(tesseractOutputFileName + ".tiff", pix, 5); // write the used image for later bold detection
        instance.setVariable("user_defined_dpi", String.valueOf(dpi));
        instance.setPageSegMode(psm);
        instance.createDocumentsWithResults(pix, null, tesseractOutputFileName, List.of(ITesseract.RenderedFormat.HOCR), ITessAPI.TessPageIteratorLevel.RIL_BLOCK);
+
    }


@@ -1,55 +0,0 @@
-package com.knecon.fforesight.service.ocr.processor.service.threads;
-
-import java.io.BufferedReader;
-import java.io.InputStream;
-import java.io.InputStreamReader;
-
-import lombok.AccessLevel;
-import lombok.AllArgsConstructor;
-import lombok.SneakyThrows;
-import lombok.experimental.FieldDefaults;
-import lombok.extern.slf4j.Slf4j;
-
-@Slf4j
-@AllArgsConstructor
-@FieldDefaults(makeFinal = true, level = AccessLevel.PRIVATE)
-public class ProcessIOLogger extends Thread {
-
-    // If the stdError or stdOut buffer of a thread is not being emptied it might lock the process in case of errors, so we need to empty both streams to prevent a deadlock.
-    // Since both need to read simultaneously we need to implement the readers as separate threads.
-
-    InputStream is;
-    String processName;
-    Type type;
-
-
-    @SneakyThrows
-    public void run() {
-
-        try (InputStreamReader isr = new InputStreamReader(is); BufferedReader br = new BufferedReader(isr)) {
-
-            String line;
-            while (true) {
-                line = br.readLine();
-
-                if (line == null) {
-                    break;
-                }
-
-                if (type.equals(Type.ERROR)) {
-                    log.error(processName + "_" + type.name() + ">" + line);
-                } else {
-                    log.debug(processName + "_" + type.name() + ">" + line);
-                }
-            }
-        }
-        is.close();
-    }
-
-
-    public enum Type {
-        ERROR,
-        STD_OUT
-    }
-
-}
@@ -14,14 +14,17 @@ public class OcrServiceSettings {

    int ocrThreadCount = 4; // Number of OCR threads
    int imageExtractThreadCount = 2; // Number of image extraction threads
-    int gsProcessCount = 2; // Number of Ghostscript processes
+    int gsProcessCount = 1; // Number of Ghostscript processes
    int dpi = 300; // Target DPI for binarized images
    int psmOverride = -1; // Overrides the page segmentation mode if > 0
    int minImageHeight = 20; // Minimum height for images to be processed
    int minImageWidth = 20; // Minimum width for images to be processed
+    float minRotationConfidence = 2; // Sets a lower bound for the confidence rating for rotated pages.
    boolean debug; // If true, overlays OCR images with a grid and draws word bounding boxes
    boolean removeWatermark; // If true, watermarks will be removed
    String languages = "deu+eng"; // Defines languages loaded into Tesseract as 3-char codes, additional languages must also be installed in the docker environment
    COSName ocrMarkedContentTag = COSName.getPDFName("KNECON_OCR");
+    boolean boldDetection = true; // if true, bold detection will be attempted
+    double boldThreshold = 0.5; // Words are opened with a brick of average stroke width, if the ratio of remaining pixels is higher the word is determined bold.

 }
@@ -0,0 +1,85 @@
+package com.knecon.fforesight.service.ocr.processor.utils;
+
+import java.awt.AlphaComposite;
+import java.awt.Color;
+import java.awt.Graphics;
+import java.awt.Graphics2D;
+import java.awt.Transparency;
+import java.awt.image.BufferedImage;
+import java.nio.IntBuffer;
+
+import org.apache.pdfbox.pdmodel.graphics.color.PDDeviceGray;
+import org.apache.pdfbox.pdmodel.graphics.color.PDDeviceRGB;
+
+import com.knecon.fforesight.service.ocr.processor.model.ExtractedImage;
+import com.sun.jna.ptr.PointerByReference;
+
+import lombok.SneakyThrows;
+import lombok.experimental.UtilityClass;
+import net.sourceforge.lept4j.L_Kernel;
+import net.sourceforge.lept4j.Leptonica1;
+import net.sourceforge.lept4j.Pix;
+import net.sourceforge.lept4j.util.LeptUtils;
+
+@UtilityClass
+public class ImageProcessingUtils {
+
+    public BufferedImage convertToDeviceColorSpace(ExtractedImage extractedImage) {
+
+        BufferedImage image;
+        if (extractedImage.colorSpace() instanceof PDDeviceRGB || extractedImage.colorSpace() instanceof PDDeviceGray) {
+            image = extractedImage.image();
+        } else {
+            BufferedImage pdfImage = extractedImage.image();
+            image = new BufferedImage(pdfImage.getWidth(), pdfImage.getHeight(), BufferedImage.TYPE_BYTE_GRAY);
+            Graphics g = image.getGraphics();
+            g.drawImage(pdfImage, 0, 0, null);
+            g.dispose();
+        }
+        return image;
+    }
+
+
+    public Pix deRotatePix(int orientDegree, Pix pix) {
+
+        return switch (360 - orientDegree) {
+            case 90 -> Leptonica1.pixRotateOrth(pix, 1);
+            case 180 -> Leptonica1.pixRotateOrth(pix, 2);
+            case 270 -> Leptonica1.pixRotateOrth(pix, 3);
+            default -> pix;
+        };
+    }
+
+
+    public static void setAlphaChannelToWhite(BufferedImage image) {
+
+        if (image.getTransparency() == Transparency.TRANSLUCENT) {
+            // NOTE: For BITMASK images, the color model is likely IndexColorModel,
+            // and this model will contain the "real" color of the transparent parts
+            // which is likely a better fit than unconditionally setting it to white.
+
+            // Fill background  with white
+            Graphics2D graphics = image.createGraphics();
+            try {
+                graphics.setComposite(AlphaComposite.DstOver); // Set composite rules to paint "behind"
+                graphics.setPaint(Color.WHITE);
+                graphics.fillRect(0, 0, image.getWidth(), image.getHeight());
+            } finally {
+                graphics.dispose();
+            }
+        }
+    }
+
+
+    public static double calculatePixelDensity(Pix pix) {
+
+        IntBuffer pixelCount = IntBuffer.allocate(1);
+        int result = Leptonica1.pixCountPixels(pix, pixelCount, null);
+        if (result == 0) {
+            return (double) pixelCount.get() / (pix.h * pix.w);
+        } else {
+            return -1;
+        }
+    }
+
+}
@@ -0,0 +1,73 @@
+package com.knecon.fforesight.service.ocr.processor.utils;
+
+import lombok.experimental.UtilityClass;
+import net.sourceforge.lept4j.L_Kernel;
+import net.sourceforge.lept4j.Leptonica1;
+
+@UtilityClass
+public class KernelUtils {
+
+    /*
+    -1, -1, -1
+    -1, 8, -1
+    -1, -1, -1
+     */
+    public L_Kernel createFullLaplacianKernel() {
+
+        L_Kernel laplacianKernel = Leptonica1.kernelCreate(3, 3);
+        Leptonica1.kernelSetElement(laplacianKernel, 0, 0, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 0, 1, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 0, 2, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 0, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 2, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 0, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 1, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 2, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 1, 8);
+        return laplacianKernel;
+    }
+
+    /*
+    0, 0, -1, 0, 0
+    0, -1, -1, -1, 0
+    -1, -1, 12, -1, -1
+    0, -1, -1, -1, 0
+    0, 0, -1, 0, 0
+     */
+    public L_Kernel createLaplacianKernel5x5() {
+
+        L_Kernel laplacianKernel = Leptonica1.kernelCreate(5, 5);
+        Leptonica1.kernelSetElement(laplacianKernel, 0, 2, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 1, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 2, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 3, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 0, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 1, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 3, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 4, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 3, 1, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 3, 2, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 3, 3, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 4, 2, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 2, 12);
+        return laplacianKernel;
+    }
+
+    /*
+    0, -1, 0
+    -1, 4, -1
+    0, -1, 0
+     */
+    public L_Kernel createLaplacianKernel() {
+
+        L_Kernel laplacianKernel = Leptonica1.kernelCreate(3, 3);
+        Leptonica1.kernelSetElement(laplacianKernel, 0, 1, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 0, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 2, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 2, 1, -1);
+        Leptonica1.kernelSetElement(laplacianKernel, 1, 1, 4);
+        return laplacianKernel;
+    }
+
+
+}
@@ -1,5 +1,25 @@
 package com.knecon.fforesight.service.ocr.processor.utils;

+import static net.sourceforge.tess4j.ITessAPI.TRUE;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPIDelete;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPIEnd;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPIGetIterator;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPIGetStringVariable;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPIMeanTextConf;
+import static net.sourceforge.tess4j.TessAPI1.TessBaseAPIProcessPage;
+import static net.sourceforge.tess4j.TessAPI1.TessDeleteResultRenderer;
+import static net.sourceforge.tess4j.TessAPI1.TessHOcrRendererCreate;
+import static net.sourceforge.tess4j.TessAPI1.TessPageIteratorBegin;
+import static net.sourceforge.tess4j.TessAPI1.TessPageIteratorBoundingBox;
+import static net.sourceforge.tess4j.TessAPI1.TessPageIteratorNext;
+import static net.sourceforge.tess4j.TessAPI1.TessResultIteratorConfidence;
+import static net.sourceforge.tess4j.TessAPI1.TessResultIteratorDelete;
+import static net.sourceforge.tess4j.TessAPI1.TessResultIteratorGetPageIterator;
+import static net.sourceforge.tess4j.TessAPI1.TessResultIteratorGetUTF8Text;
+import static net.sourceforge.tess4j.TessAPI1.TessResultRendererBeginDocument;
+import static net.sourceforge.tess4j.TessAPI1.TessResultRendererEndDocument;
+import static net.sourceforge.tess4j.TessAPI1.TessResultRendererInsert;
+
 import java.awt.Rectangle;
 import java.nio.IntBuffer;
 import java.util.ArrayList;
@@ -9,17 +29,19 @@ import com.sun.jna.Pointer;

 import lombok.extern.slf4j.Slf4j;
 import net.sourceforge.lept4j.Pix;
+import net.sourceforge.tess4j.ITessAPI;
 import net.sourceforge.tess4j.OCRResult;
 import net.sourceforge.tess4j.TessAPI1;
-import net.sourceforge.tess4j.Tesseract1;
+import net.sourceforge.tess4j.Tesseract;
 import net.sourceforge.tess4j.TesseractException;
 import net.sourceforge.tess4j.Word;

@Slf4j
-public class Tesseract2 extends Tesseract1 {
+/**
+ * Overriden version only so I can use Tesseract1 with Pixs instead of BufferedImages. All Functions are copied and then the BufferedImage -> Pix conversion deleted.
+ */ public class Tesseract2 extends Tesseract {

-
-    private int createDocuments(Pix pix, String filename, TessResultRenderer renderer) {
+    private int createDocuments(Pix pix, String filename, ITessAPI.TessResultRenderer renderer) {

        String title = TessBaseAPIGetStringVariable(getHandle(), DOCUMENT_TITLE);
        TessResultRendererBeginDocument(renderer, title);
@@ -59,7 +81,7 @@ public class Tesseract2 extends Tesseract1 {
        try {
            for (int i = 0; i < pixs.length; i++) {
                try {
-                    TessResultRenderer renderer = createRenderers(outputbases[i], formats);
+                    ITessAPI.TessResultRenderer renderer = createRenderers(outputbases[i], formats);
                    int meanTextConfidence = createDocuments(pixs[i], filenames[i], renderer);
                    TessDeleteResultRenderer(renderer);
                    List<Word> words = meanTextConfidence > 0 ? getRecognizedWords(pageIteratorLevel) : new ArrayList<Word>();
@@ -82,8 +104,8 @@ public class Tesseract2 extends Tesseract1 {
        List<Word> words = new ArrayList<>();

        try {
-            TessResultIterator ri = TessBaseAPIGetIterator(getHandle());
-            TessPageIterator pi = TessResultIteratorGetPageIterator(ri);
+            ITessAPI.TessResultIterator ri = TessBaseAPIGetIterator(getHandle());
+            ITessAPI.TessPageIterator pi = TessResultIteratorGetPageIterator(ri);
            TessPageIteratorBegin(pi);

            do {
@@ -116,9 +138,9 @@ public class Tesseract2 extends Tesseract1 {
    }


-    private TessResultRenderer createRenderers(String outputbase, List<RenderedFormat> formats) {
+    private ITessAPI.TessResultRenderer createRenderers(String outputbase, List<RenderedFormat> formats) {

-        TessResultRenderer renderer = null;
+        ITessAPI.TessResultRenderer renderer = null;

        for (RenderedFormat format : formats) {
            switch (format) {
@@ -135,4 +157,12 @@ public class Tesseract2 extends Tesseract1 {
        return renderer;
    }

+
+    @Override
+    protected void dispose() {
+
+        TessBaseAPIEnd(getHandle());
+        TessBaseAPIDelete(getHandle());
+    }
+
 }
@@ -20,7 +20,7 @@ class Type0FontMetricsFactoryTest {
    public void testStringWidth() {

        try (PDDocument document = Loader.loadPDF(new File(Type0FontMetricsFactoryTest.class.getClassLoader().getResource("InvisibleText.pdf").getPath()))) {
-            Type0FontMetricsFactory metricsFactory = new Type0FontMetricsFactory(document);
+            Type0FontMetricsFactory metricsFactory = Type0FontMetricsFactory.regular(document);
            FontMetrics fontMetrics = metricsFactory.calculateMetrics("deine mutter", 100, 50);
        }

@@ -0,0 +1,36 @@
+package com.knecon.fforesight.service.ocr.processor.utils;
+
+import static net.sourceforge.lept4j.ILeptonica.IFF_PNG;
+
+import org.junit.jupiter.api.BeforeEach;
+import org.junit.jupiter.api.Disabled;
+import org.junit.jupiter.api.Test;
+
+import net.sourceforge.lept4j.Leptonica1;
+import net.sourceforge.lept4j.Pix;
+
+@Disabled
+class ImageProcessingUtilsTest {
+
+    @BeforeEach
+    public void loadLeptonica() {
+
+        System.setProperty("jna.library.path", System.getenv("VCPKG_DYNAMIC_LIB"));
+    }
+
+
+    @Test
+    public void testRotation() {
+
+        Pix pix = Leptonica1.pixRead("/home/kschuettler/Downloads/painHarold.webp");
+        Pix pix2 = ImageProcessingUtils.deRotatePix(0, pix);
+        Leptonica1.pixWrite("/tmp/0.png", pix2, IFF_PNG);
+        Pix pix3 = ImageProcessingUtils.deRotatePix(90, pix);
+        Leptonica1.pixWrite("/tmp/90.png", pix3, IFF_PNG);
+        Pix pix4 = ImageProcessingUtils.deRotatePix(180, pix);
+        Leptonica1.pixWrite("/tmp/180.png", pix4, IFF_PNG);
+        Pix pix5 = ImageProcessingUtils.deRotatePix(270, pix);
+        Leptonica1.pixWrite("/tmp/270.png", pix5, IFF_PNG);
+    }
+
+}
@@ -1,10 +1,7 @@
 package com.knecon.fforesight.service.ocr.processor.utils;

 import java.awt.image.BufferedImage;
-import java.io.BufferedReader;
 import java.io.File;
-import java.io.InputStream;
-import java.io.InputStreamReader;
 import java.util.LinkedList;
 import java.util.List;
 import java.util.stream.IntStream;
@@ -19,7 +16,7 @@ import org.springframework.core.io.ClassPathResource;
 import org.springframework.util.FileSystemUtils;

 import com.knecon.fforesight.service.ocr.processor.service.OsUtils;
-import com.knecon.fforesight.service.ocr.processor.service.threads.ProcessIOLogger;
+import com.knecon.fforesight.service.ocr.processor.service.threads.GhostScriptOutputHandler;

 import lombok.SneakyThrows;

@@ -50,29 +47,6 @@ public class Pdf2ImgTest {
    }


-    @Test
-    @SneakyThrows
-    public void testGhostScript() {
-
-        String outputDir = "/tmp/ghostscript_out/";
-        new File(outputDir).mkdirs();
-        ClassPathResource resource = new ClassPathResource("files/Cyberport__SD-Faktura-Kopie_(ZRG2)_-_31.08.2020.pdf");
-
-        String[] cmdArgs = new String[]{"gs", "-dNOPAUSE", "-sDEVICE=tiff24nc", "-r" + DPI, "-sOutputFile=" + outputDir + "page%04d", resource.getFile().toString(), "-c", "quit"};
-        Process p = Runtime.getRuntime().exec(cmdArgs);
-        ProcessIOLogger logger = new ProcessIOLogger(p.getInputStream(), "GS", ProcessIOLogger.Type.STD_OUT);
-        logger.start();
-        ProcessIOLogger errorLogger = new ProcessIOLogger(p.getErrorStream(), "GS", ProcessIOLogger.Type.STD_OUT);
-        errorLogger.start();
-        int exitcode = p.waitFor();
-        logger.join();
-        errorLogger.join();
-        System.out.println("Ghostscript finished with exit code " + exitcode);
-        FileSystemUtils.deleteRecursively(new File(outputDir));
-
-    }
-
-
    @Test
    @SneakyThrows
    public void testGhostScriptParallel() {
@@ -3,7 +3,7 @@ import org.springframework.boot.gradle.tasks.bundling.BootBuildImage
 plugins {
    application
    id("com.iqser.red.service.java-conventions")
-    id("org.springframework.boot") version "3.1.3"
+    id("org.springframework.boot") version "3.1.5"
    id("io.spring.dependency-management") version "1.1.3"
    id("org.sonarqube") version "4.3.0.3225"
    id("io.freefair.lombok") version "8.2.2"
@@ -11,19 +11,26 @@ plugins {

 configurations {
    all {
-        exclude(group = "org.springframework.boot", module = "spring-boot-starter-logging")
+        exclude(group = "commons-logging", module = "commons-logging")
+        exclude(group = "org.springframework.boot", module = "spring-boot-starter-log4j2")
+        exclude(group = "com.iqser.red.commons", module = "logging-commons")
    }
 }

+val springBootStarterVersion = "3.1.5"

 dependencies {
    implementation(project(":ocr-service-processor"))
    implementation(project(":ocr-service-api"))

+    implementation("com.knecon.fforesight:tracing-commons:0.3.0")
    implementation("org.springframework.cloud:spring-cloud-starter-openfeign:4.0.4")
-    implementation("org.springframework.boot:spring-boot-starter-amqp:3.1.4")
+    implementation("org.springframework.boot:spring-boot-starter-amqp:${springBootStarterVersion}")

-    testImplementation("org.springframework.boot:spring-boot-starter-test:3.1.4")
+    implementation("net.logstash.logback:logstash-logback-encoder:7.4")
+    implementation("ch.qos.logback:logback-classic")
+
+    testImplementation("org.springframework.boot:spring-boot-starter-test:${springBootStarterVersion}")
    testImplementation("com.iqser.red.commons:test-commons:2.1.0")
    testImplementation("org.springframework.amqp:spring-rabbit-test:3.0.2")
 }
@@ -5,10 +5,16 @@ persistence-service.url: "http://persistence-service-v1:8080"
 tenant-user-management-service.url: "http://tenant-user-management-service:8080/internal"
 fforesight.tenants.remote: true

+logging.type: ${LOGGING_TYPE:CONSOLE}
+kubernetes.namespace: ${NAMESPACE:default}
+project.version: 1.0-SNAPSHOT
+
 server:
  port: 8080

 spring:
+  application:
+    name: ocr-service
  main:
    allow-circular-references: true # FIXME
  profiles:
@@ -33,6 +39,7 @@ fforesight:
    ignored-endpoints: [ '/actuator/health', '/actuator/health/**' ]
    enabled: true

+logging.pattern.level: "%5p [${spring.application.name},%X{traceId:-},%X{spanId:-}]"

 management:
  endpoint:
@@ -41,9 +48,12 @@ management:
    health.enabled: true
  endpoints.web.exposure.include: prometheus, health, metrics
  metrics.export.prometheus.enabled: ${monitoring.enabled:false}
-
-
-storage:
-  backend: 's3'
+  tracing:
+    enabled: ${TRACING_ENABLED:false}
+    sampling:
+      probability: ${TRACING_PROBABILITY:1.0}
+  otlp:
+    tracing:
+      endpoint: ${OTLP_ENDPOINT:http://otel-collector-opentelemetry-collector.otel-collector:4318/v1/traces}

 pdftron.license: ${PDFTRON_LICENSE}
@@ -0,0 +1,17 @@
+<configuration>
+
+    <springProperty scope="configuration" name="logType" source="logging.type"/>
+    <springProperty scope="context" name="application.name" source="spring.application.name"/>
+    <springProperty scope="context" name="version" source="project.version"/>
+    <include resource="org/springframework/boot/logging/logback/defaults.xml"/>
+    <include resource="org/springframework/boot/logging/logback/console-appender.xml"/>
+
+    <appender name="JSON" class="ch.qos.logback.core.ConsoleAppender">
+        <encoder class="net.logstash.logback.encoder.LogstashEncoder"/>
+    </appender>
+
+    <root level="INFO">
+        <appender-ref ref="${logType}"/>
+    </root>
+
+</configuration>
@@ -4,10 +4,15 @@ import static com.iqser.red.pdftronlogic.commons.PdfTextExtraction.extractAllTex
 import static com.knecon.fforesight.service.ocr.processor.service.OsUtils.getTemporaryDirectory;
 import static org.assertj.core.api.Assertions.assertThat;

+import java.io.File;
 import java.io.FileInputStream;
 import java.io.FileOutputStream;
+import java.nio.file.Files;
 import java.nio.file.Path;
+import java.util.Comparator;
+import java.util.List;
 import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicInteger;

 import org.junit.jupiter.api.BeforeEach;
 import org.junit.jupiter.api.Disabled;
@@ -59,7 +64,7 @@ public class OcrServiceIntegrationTest extends AbstractTest {
    @SneakyThrows
    public void testOcr() {

-        String text = testOCR("files/2009-1048395_50pages_tables.pdf");
+        String text = testOCR("files/402Study.pdf");
    }


@@ -127,4 +132,55 @@ public class OcrServiceIntegrationTest extends AbstractTest {
        }
    }

+
+    @Test
+    @SneakyThrows
+    public void testOcrForAllDMFiles() {
+
+        String dir = "/home/kschuettler/Dokumente/TestFiles/syn-dm-testfiles/";
+        List<File> foundFiles = Files.walk(Path.of(dir))
+                .sorted(Comparator.comparingLong(this::getFileSize))
+                .map(Path::toFile)
+                .filter(file -> file.getName().endsWith(".pdf"))
+                .peek(System.out::println)
+                .toList();
+        int fileCount = foundFiles.size();
+        AtomicInteger processedCount = new AtomicInteger();
+        System.out.printf("Found %s files, starting OCR for each.%n%n", fileCount);
+        foundFiles.stream().peek(file -> System.out.printf("%s/%s: %s%n", processedCount.getAndIncrement(), fileCount, file)).forEach(this::testOCRForFile);
+    }
+
+
+    @SneakyThrows
+    public long getFileSize(Path path) {
+
+        return Files.size(path);
+    }
+
+
+    @Test
+    @SneakyThrows
+    public void testOcrForSpecificFile() {
+
+        testOCRForFile(new File("/home/kschuettler/Dokumente/TestFiles/syn-dm-testfiles2/A16361B - Acute Dermal Toxicity Study in Rats.pdf"));
+    }
+
+
+
+    @SneakyThrows
+    private void testOCRForFile(File file) {
+
+        var originId = FileStorageService.getStorageId(TEST_DOSSIER_ID, "file", FileType.ORIGIN);
+        try (var fileStream = new FileInputStream(file)) {
+            storageService.storeObject(TenantContext.getTenantId(), originId, fileStream);
+        }
+
+        Path tmpFileName = Path.of(getTemporaryDirectory()).resolve(Path.of(file.getAbsolutePath()).getFileName());
+        try (var out = new FileOutputStream(tmpFileName.toFile())) {
+            ocrService.runOcrOnDocument(TEST_DOSSIER_ID, "file", out);
+            System.out.println("File:" + tmpFileName);
+        }
+        System.out.println("\n\n");
+    }
+
 }
Author	SHA1	Message	Date
Dominique Eifländer	8924e905ad	hotfix: Extend Tesseract instead of Tesseract1	2024-01-15 16:13:43 +01:00
Kilian Schüttler	fb1fe35bc1	Merge branch 'RED-8155' into 'master' RED-8155: bold-detection in ocr-service Closes RED-8155 See merge request redactmanager/ocr-service!33	2024-01-08 13:53:30 +01:00
Kilian Schuettler	912f00aa84	RED-8155: bold-detection in ocr-service * fix application.yml	2024-01-08 13:49:58 +01:00
Kilian Schüttler	bab16ad9b2	Merge branch 'RED-7669-fontstyle' into 'master' RED-8155: integrate bold-detection into ocr-service Closes RED-7669 See merge request redactmanager/ocr-service!31	2024-01-05 16:05:53 +01:00
Kilian Schüttler	be4656189b	RED-8155: integrate bold-detection into ocr-service	2024-01-05 16:05:53 +01:00
Dominique Eifländer	8944b57344	Merge branch 'RED-7669' into 'master' RED-7669: optimize OCR-module performance Closes RED-7669 See merge request redactmanager/ocr-service!30	2023-12-22 15:14:42 +01:00
Kilian Schuettler	67540950b8	RED-7669: optimize OCR-module performance * fix thread handling for PDFs without any images	2023-12-22 15:11:29 +01:00
Kilian Schuettler	6f29270e66	RED-7669: optimize OCR-module performance * fix thread handling for PDFs without any images	2023-12-22 15:04:52 +01:00
Dominique Eifländer	b961c9e324	Merge branch 'RED-1137' into 'master' RED-1137: Do not observe actuator endpoints Closes RED-1137 See merge request redactmanager/ocr-service!29	2023-12-20 14:24:49 +01:00
Dominique Eifländer	4b6411161e	RED-1137: Do not observe actuator endpoints	2023-12-20 14:17:09 +01:00
Dominique Eifländer	14982eae7c	Merge branch 'RED-5223' into 'master' RED-5223: Use tracing-commons from fforesight Closes RED-5223 See merge request redactmanager/ocr-service!28	2023-12-13 16:16:29 +01:00
Dominique Eifländer	99fc16130b	RED-5223: Use tracing-commons from fforesight	2023-12-13 16:10:10 +01:00
Dominique Eifländer	80d38fb785	Merge branch 'RED-7669' into 'master' RED-7669: optimize OCR-module performance Closes RED-7669 See merge request redactmanager/ocr-service!27	2023-12-12 15:27:00 +01:00
Kilian Schüttler	c06974ce69	RED-7669: optimize OCR-module performance	2023-12-12 15:27:00 +01:00
Dominique Eifländer	591c7d7fab	Merge branch 'RED-5223' into 'master' RED-5223: Enabled tracing, upgrade spring, use logstash-logback-encoder for json logs Closes RED-5223 See merge request redactmanager/ocr-service!26	2023-12-12 12:36:01 +01:00
Dominique Eifländer	0300a087d4	RED-5223: Enabled tracing, upgrade spring, use logstash-logback-encoder for json logs	2023-12-12 11:55:01 +01:00
Andrei Isvoran	98752ff1d1	Merge branch 'RED-7714' into 'master' RED-7715 - Add log4j config to enable switching between json/line logs Closes RED-7714 See merge request redactmanager/ocr-service!25	2023-12-06 12:42:33 +01:00
Andrei Isvoran	ae09a59a7c	RED-7715 - Add log4j config to enable switching between json/line logs	2023-12-06 11:52:01 +02:00
Kilian Schüttler	65d818200f	Merge branch 'RED-7669' into 'master' RED-7669: optimize OCR-module performance Closes RED-7669 See merge request redactmanager/ocr-service!24	2023-11-28 12:35:22 +01:00
Kilian Schuettler	6fe95c6940	RED-7669: optimize OCR-module performance * dont interrupt threads, use boolean flag instead	2023-11-28 10:04:56 +01:00
Kilian Schüttler	202132e14c	Merge branch 'RED-7668' into 'master' RED-7669: optimize OCR-module performance Closes RED-7668 See merge request redactmanager/ocr-service!23	2023-11-24 10:57:45 +01:00
Kilian Schuettler	0264e28cc2	RED-7669: optimize OCR-module performance * enable caches	2023-11-24 10:21:55 +01:00
Dominique Eifländer	a50f54676e	Merge branch 'RED-7668' into 'master' RED-7669: optimize OCR-module performance Closes RED-7668 See merge request redactmanager/ocr-service!22	2023-11-23 16:04:43 +01:00
Kilian Schuettler	1926707ae1	RED-7669: optimize OCR-module performance * move all critical stuff to its own singleton thread * make gs process queue any image once the file has been written	2023-11-23 16:00:53 +01:00
Kilian Schuettler	d3190844a3	RED-7669: optimize OCR-module performance * move all critical stuff to its own singleton thread * make gs process queue any image once the file has been written	2023-11-23 16:00:31 +01:00
Kilian Schuettler	c7ccbae6ff	RED-7669: optimize OCR-module performance * move all critical stuff to its own singleton thread * make gs process queue any image once the file has been written	2023-11-23 16:00:31 +01:00
Kilian Schuettler	880bebcafc	RED-7669: optimize OCR-module performance * move all critical stuff to its own singleton thread * make gs process queue any image once the file has been written	2023-11-23 16:00:31 +01:00
Kilian Schuettler	955ff6281d	RED-7669: optimize OCR-module performance * move all critical stuff to its own singleton thread * make gs process queue any image once the file has been written	2023-11-23 16:00:31 +01:00
Kilian Schuettler	efd3a1d952	RED-7669: optimize OCR-module performance * move all non thread safe stuff to separate thread in the middle	2023-11-23 16:00:29 +01:00
Kilian Schuettler	bb5b4a2fd8	RED-7669: optimize OCR-module performance * binarize images after reading	2023-11-23 16:00:22 +01:00
Kilian Schuettler	6f99664906	RED-7669: optimize OCR-module performance * try and synchronize all malloc calls	2023-11-23 16:00:19 +01:00
Kilian Schuettler	574f7ac25e	RED-7669: optimize OCR-module performance * moar sigsegv	2023-11-23 16:00:01 +01:00
Kilian Schuettler	12217f2459	RED-7669: optimize OCR-module performance * moar sigsegv	2023-11-23 16:00:01 +01:00
Kilian Schuettler	19747cbca5	RED-7669: optimize OCR-module performance * moar sigsegv	2023-11-23 15:59:59 +01:00
Kilian Schuettler	2632d2023d	RED-7669: optimize OCR-module performance * reset test and settings	2023-11-23 15:59:16 +01:00
Kilian Schuettler	4c225c2219	RED-7669: optimize OCR-module performance * cleanup Code	2023-11-23 15:59:16 +01:00
Kilian Schuettler	3d09f46844	RED-7669: optimize OCR-module performance * don't despeckle small images	2023-11-23 15:59:16 +01:00
Kilian Schuettler	77355b5367	RED-7669: optimize OCR-module performance * second attempt at thread safety	2023-11-23 15:59:16 +01:00
Kilian Schuettler	57e194fcd0	RED-7669: optimize OCR-module performance * attempt at thread safety	2023-11-23 15:59:14 +01:00
Kilian Schüttler	c556687499	Merge branch 'RED-7669' into 'master' RED-7669: optimize OCR-module performance Closes RED-7669 See merge request redactmanager/ocr-service!21	2023-11-20 09:55:48 +01:00
Kilian Schüttler	759bae6499	RED-7669: optimize OCR-module performance	2023-11-20 09:55:48 +01:00