Merge branch 'memory-optimisations' into 'master'

Memory optimisations

See merge request redactmanager/commons/pdftron-logic-commons!32
This commit is contained in:
Timo Bejan 2024-03-15 08:55:34 +01:00
commit 16ef5afe90
6 changed files with 283 additions and 199 deletions

View File

@ -1,7 +1,5 @@
package com.iqser.red.pdftronlogic.commons;
import static com.iqser.red.pdftronlogic.commons.InvisibleElementRemovalService.TOLERANCE;
import java.awt.geom.Area;
@ -35,6 +33,13 @@ public class ClippingPathStack {
}
@SneakyThrows
public void intersectClippingPath(Rectangle2D path) {
getCurrentClippingPath().intersect(new Area(path));
}
public boolean almostIntersects(double x, double y, double width, double height) {
// To address inconsistencies in the calculation of the bounding box we slightly increase the rectangle
// Height or width are zero for straight lines, even though they are being rendered. Therefore, height or width must be at minimum >0.
@ -56,15 +61,16 @@ public class ClippingPathStack {
public void enterNewGState() {
Area current = stack.peek();
Area cloned = new Area();
cloned.add(current);
Area cloned = (Area) current.clone();
stack.push(cloned);
}
public void leaveGState() {
stack.pop();
// somehow this greatly helps memory management
var popped = stack.pop();
popped.reset();
}
}

View File

@ -4,6 +4,7 @@ import com.pdftron.common.PDFNetException;
import com.pdftron.pdf.Element;
public class ElementFeatureFactory {
public static ElementFeatures extractFeatures(Element element) throws PDFNetException {
return switch (element.getType()) {
@ -16,7 +17,9 @@ public class ElementFeatureFactory {
};
}
public static ElementFeatures extractFeaturesWithHash(Element element, String hashObject) throws PDFNetException {
return buildImage(element)
.hashOfImage(hashObject)
.build();
@ -25,20 +28,23 @@ public class ElementFeatureFactory {
private static ElementFeatures.Form buildForm(Element element) throws PDFNetException {
try (var bbox = element.getBBox();) {
return ElementFeatures.Form.builder()
.elementType(element.getType())
.boundingBox(Converter.toRectangle2D(element.getBBox()))
.boundingBox(Converter.toRectangle2D(bbox))
.xObjectType(element.getXObject().getType())
.dictOrArrayOrStreamLength(element.getXObject().getType() == 7 ? element.getXObject().getDecodedStream().size() : 0)
.build();
}
}
private static ElementFeatures.Image.ImageBuilder<?, ?> buildImage(Element element) throws PDFNetException {
try (var bbox = element.getBBox();) {
return ElementFeatures.Image.builder()
.elementType(element.getType())
.boundingBox(Converter.toRectangle2D(element.getBBox()))
.boundingBox(Converter.toRectangle2D(bbox))
.dataSize(element.getImageDataSize())
.height(element.getImageHeight())
.width(element.getImageWidth())
@ -46,35 +52,41 @@ public class ElementFeatureFactory {
.componentNum(element.getComponentNum())
.bitsPerComponent(element.getBitsPerComponent());
}
}
private static ElementFeatures.Text buildText(Element element) throws PDFNetException {
try (var bbox = element.getBBox();) {
return ElementFeatures.Text.builder()
.elementType(element.getType())
.boundingBox(Converter.toRectangle2D(element.getBBox()))
.boundingBox(Converter.toRectangle2D(bbox))
.text(element.getTextString())
.font(element.getGState().getFont().getType())
.fontsize(element.getGState().getFontSize())
.build();
}
}
private static ElementFeatures.Path buildPath(Element element) throws PDFNetException {
try (var bbox = element.getBBox(); var ctm = element.getCTM();
var fillColor = element.getGState().getFillColor();
var strokeColor = element.getGState().getStrokeColor()) {
return ElementFeatures.Path.builder()
.elementType(element.getType())
.boundingBox(Converter.toRectangle2D(element.getBBox()))
.boundingBox(Converter.toRectangle2D(bbox))
.isClippingPath(element.isClippingPath())
.isClipWindingFill(element.isClipWindingFill())
.isStroked(element.isStroked())
.isFilled(element.isFilled())
.isWindingFill(element.isWindingFill())
.fillColor(Converter.convertColor(element.getGState().getFillColorSpace(), element.getGState().getFillColor()))
.strokeColor(Converter.convertColor(element.getGState().getStrokeColorSpace(), element.getGState().getStrokeColor()))
.linePath(Converter.convertToGeneralPathAndTransformToInitialUserSpace(element.getPathData(), element.getCTM()))
.fillColor(Converter.convertColor(element.getGState().getFillColorSpace(), fillColor))
.strokeColor(Converter.convertColor(element.getGState().getStrokeColorSpace(), strokeColor))
.linePath(Converter.convertToGeneralPathAndTransformToInitialUserSpace(element.getPathData(), ctm))
.build();
}
}
}

View File

@ -31,9 +31,11 @@ public class ElementFeatures {
public boolean almostMatches(Element element) throws PDFNetException {
try (var bbox = element.getBBox()) {
return element.getType() == elementType && //
element.getBBox() != null && //
rectsAlmostMatch(element.getBBox());
bbox != null && //
rectsAlmostMatch(bbox);
}
}
@ -199,7 +201,10 @@ public class ElementFeatures {
if (elementFeatures.getClass() != this.getClass()) {
return false;
}
return super.almostMatches(elementFeatures) && this.dataSize == ((Image) elementFeatures).getDataSize() && this.height == ((Image) elementFeatures).getHeight() && this.width == ((Image) elementFeatures).getWidth() && this.renderingIntent == ((Image) elementFeatures).getRenderingIntent() && this.componentNum == ((Image) elementFeatures).getComponentNum() && this.bitsPerComponent == ((Image) elementFeatures).getBitsPerComponent() && calculateHammingDistance(
return super.almostMatches(elementFeatures) && this.dataSize == ((Image) elementFeatures).getDataSize() && this.height == ((Image) elementFeatures).getHeight() &&
this.width == ((Image) elementFeatures).getWidth() && this.renderingIntent == ((Image) elementFeatures).getRenderingIntent() &&
this.componentNum == ((Image) elementFeatures).getComponentNum() && this.bitsPerComponent == ((Image) elementFeatures).getBitsPerComponent() &&
calculateHammingDistance(
((Image) elementFeatures).getHashOfImage()) <= HAMMING_DISTANCE_THRESHOLD;
}
@ -241,11 +246,13 @@ public class ElementFeatures {
@Override
public boolean almostMatches(Element element) throws PDFNetException {
try (var bbox = element.getBBox()) {
return element.getType() == getElementType() && //
element.getBBox() != null && //
(super.rectsAlmostMatch(element.getBBox()) || almostRotateMatches(element.getBBox().getRectangle())) && xObjectType == element.getXObject()
bbox != null && //
(super.rectsAlmostMatch(bbox) || almostRotateMatches(bbox.getRectangle())) && xObjectType == element.getXObject()
.getType() && dictOrArrayOrStreamLength == element.getXObject().getDecodedStream().size();
}
}
public boolean almostMatches(ElementFeatures elementFeatures) {
@ -253,9 +260,11 @@ public class ElementFeatures {
if (elementFeatures.getClass() != this.getClass()) {
return false;
}
return elementFeatures.getElementType() == getElementType() && elementFeatures.getBoundingBox() != null && (super.rectsAlmostMatch(elementFeatures.getBoundingBox()) || almostRotateMatches(
return elementFeatures.getElementType() == getElementType() && elementFeatures.getBoundingBox() != null &&
(super.rectsAlmostMatch(elementFeatures.getBoundingBox()) || almostRotateMatches(
elementFeatures.getBoundingBox()
.getBounds2D())) && xObjectType == ((Form) elementFeatures).getXObjectType() && dictOrArrayOrStreamLength == ((Form) elementFeatures).getDictOrArrayOrStreamLength();
.getBounds2D())) && xObjectType == ((Form) elementFeatures).getXObjectType() &&
dictOrArrayOrStreamLength == ((Form) elementFeatures).getDictOrArrayOrStreamLength();
}

View File

@ -63,15 +63,12 @@ public class InvisibleElementRemovalService {
PDFDoc pdfDoc = new PDFDoc(pdfFile);
try (pdfDoc) {
execute(pdfDoc, delta, removePaths, markedContentToIgnore);
try {
pdfDoc.save(out, SDFDoc.SaveMode.LINEARIZED, null);
} catch (Exception e) {
log.error("File could not be saved after invisible element removal");
throw new RuntimeException(e);
} finally {
pdfDoc.close();
}
}
@ -289,13 +286,17 @@ public class InvisibleElementRemovalService {
if (!inClippingPath) {
gState.setFillColorSpace(ColorSpace.createDeviceRGB());
// red for elements removed by clipping path
gState.setFillColor(new ColorPt(1, 0, 0));
try (var color = new ColorPt(1, 0, 0)) {
gState.setFillColor(color);
}
writer.writeElement(textElement);
}
if (!isTextVisible) {
gState.setFillColorSpace(ColorSpace.createDeviceRGB());
// blue for elements removed due to transparency or not rendered or same color as background
gState.setFillColor(new ColorPt(0, 0, 1));
try (var color = new ColorPt(0, 0, 1)) {
gState.setFillColor(color);
}
gState.setTextRenderMode(GState.e_fill_text);
gState.setFillOpacity(1);
writer.writeElement(textElement);
@ -316,7 +317,8 @@ public class InvisibleElementRemovalService {
// see ElementEditTest in PDFTron (https://www.pdftron.com/documentation/samples/android/java/ElementEditTest)
try (ElementWriter formWriter = new ElementWriter()) {
context.clippingPathStack().enterNewGState();
context.clippingPathStack().intersectClippingPath(new GeneralPath(Converter.toRectangle2D(formElement.getBBox())));
try (var formElementBBOX = formElement.getBBox()) {
context.clippingPathStack().intersectClippingPath(Converter.toRectangle2D(formElementBBOX));
context.reader().formBegin();
formWriter.begin(formObj);
@ -330,18 +332,21 @@ public class InvisibleElementRemovalService {
}
}
}
}
private void processPath(Element pathElement, ElementWriter writer, InvisibleElementRemovalContext context) throws PDFNetException {
PathData pathData = pathElement.getPathData();
if (pathData.getOperators().length == 0 && pathData.getPoints().length == 0 || pathElement.getBBox() == null) {
try (var bbox = pathElement.getBBox()) {
if (pathData.getOperators().length == 0 && pathData.getPoints().length == 0 || bbox == null) {
writer.writeElement(pathElement);
return;
}
GeneralPath linePath = Converter.convertToGeneralPathAndTransformToInitialUserSpace(pathData, pathElement.getCTM());
try (var ctm = pathElement.getCTM()) {
GeneralPath linePath = Converter.convertToGeneralPathAndTransformToInitialUserSpace(pathData, ctm);
var rect = linePath.getBounds2D();
@ -377,14 +382,18 @@ public class InvisibleElementRemovalService {
}
if (context.delta() && !inClippingPath && context.removePaths()) {
try (var color = new ColorPt(1, 0, 0)) {
pathElement.getGState().setFillColorSpace(ColorSpace.createDeviceRGB());
pathElement.getGState().setFillColor(new ColorPt(1, 0, 0));
pathElement.getGState().setFillColor(color);
pathElement.getGState().setStrokeColorSpace(ColorSpace.createDeviceRGB());
pathElement.getGState().setStrokeColor(new ColorPt(1, 0, 0));
pathElement.getGState().setStrokeColor(color);
writer.writeElement(pathElement);
}
}
}
}
}
}
private void calculateOverlapsForLinePath(InvisibleElementRemovalContext context, GeneralPath linePath) {
@ -516,7 +525,9 @@ public class InvisibleElementRemovalService {
private boolean fillIsVisible(GState gState, Rect textBBox, InvisibleElementRemovalContext context) throws PDFNetException {
return gState.getFillOpacity() != 0 && differentColorThanBackgroundColor(Converter.convertColor(gState.getFillColorSpace(), gState.getFillColor()), textBBox, context);
try (var color = gState.getFillColor()) {
return gState.getFillOpacity() != 0 && differentColorThanBackgroundColor(Converter.convertColor(gState.getFillColorSpace(), color), textBBox, context);
}
}
@ -541,13 +552,16 @@ public class InvisibleElementRemovalService {
private static List<ElementFeatures.Path> findVisiblePathElementsThatIntersect(Rect textBBox, InvisibleElementRemovalContext context) {
return context.visibleElements()
.stream()
.filter(elementFeatures -> elementFeatures.getElementType() == Element.e_path)
.map(elementFeatures -> (ElementFeatures.Path) elementFeatures)
.filter(elementFeatures -> !elementFeatures.getFillColor().equals(Color.WHITE))
.filter(element -> element.isBackground(textBBox))
.toList();
var result = new ArrayList<ElementFeatures.Path>();
for (var element : context.visibleElements()) {
if (element.getElementType() == Element.e_path
&& !((ElementFeatures.Path) element).getFillColor().equals(Color.WHITE)
&& ((ElementFeatures.Path) element).isBackground(textBBox)) {
result.add((ElementFeatures.Path) element);
}
}
return result;
}

View File

@ -123,7 +123,8 @@ public class WatermarkRemovalService {
double minAreaCoveringPage,
Page page) throws PDFNetException {
if (element.getBBox() == null) {
try (var bbox = element.getBBox()) {
if (bbox == null) {
return;
}
@ -132,6 +133,7 @@ public class WatermarkRemovalService {
case Element.e_image, Element.e_inline_image -> processImages(element, elementFeaturesLinkedList, page, minAreaCoveringPage);
case Element.e_text -> processText(element, elementFeaturesLinkedList, page);
}
}
}
@ -147,12 +149,14 @@ public class WatermarkRemovalService {
return;
}
boolean isBigEnough = Math.abs(element.getBBox().getY1() - element.getBBox().getY2()) > page.getPageHeight() * MIN_TEXTWATERMARK_HEIGHT_THRESHOLD;
try (var bbox = element.getBBox()) {
boolean isBigEnough = Math.abs(bbox.getY1() - bbox.getY2()) > page.getPageHeight() * MIN_TEXTWATERMARK_HEIGHT_THRESHOLD;
if (isBigEnough) {
ElementFeatures elementFeatures = ElementFeatureFactory.extractFeatures(element);
elementFeaturesLinkedList.add(elementFeatures);
}
}
}
@ -160,9 +164,11 @@ public class WatermarkRemovalService {
@SneakyThrows
private boolean isTextRotated(Element element) {
return Math.abs(element.getCTM().getB()) < Math.sin(Math.toRadians(ROTATED_TEXT_THRESHOLD)) || Math.abs(element.getCTM()
try (var ctm = element.getCTM()) {
return Math.abs(ctm.getB()) < Math.sin(Math.toRadians(ROTATED_TEXT_THRESHOLD)) || Math.abs(ctm
.getB()) > Math.sin(Math.toRadians(70 - ROTATED_TEXT_THRESHOLD));
}
}
private void shouldTextSearchBeContinued(List<ElementFeatures> elementFeaturesLinkedList) {
@ -185,7 +191,8 @@ public class WatermarkRemovalService {
if (element.getXObject() == null) {
return;
}
if (element.getBBox().getHeight() * element.getBBox().getWidth() < minAreaCoveringPage && isLocatedNearBorder(element, page)) {
try (var bbox = element.getBBox()) {
if (bbox.getHeight() * bbox.getWidth() < minAreaCoveringPage && isLocatedNearBorder(element, page)) {
return;
}
@ -193,16 +200,25 @@ public class WatermarkRemovalService {
ElementFeatures elementFeatures = ElementFeatureFactory.extractFeaturesWithHash(element, hashOfImage);
elementFeaturesLinkedList.add(elementFeatures);
}
}
// Typically company logos on dossier pages are located near the border and should be excluded from the watermark removal
@SneakyThrows
private boolean isLocatedNearBorder(Element element, Page page) {
return element.getBBox().getY1() < page.getVisibleContentBox().getY1() + page.getPageHeight() * IMAGE_POSITION_HEIGHT_THRESHOLD || element.getBBox()
.getY2() > page.getVisibleContentBox().getY2() - page.getPageHeight() * IMAGE_POSITION_HEIGHT_THRESHOLD || element.getBBox().getX1() < page.getVisibleContentBox()
.getX1() + page.getPageWidth() * IMAGE_POSITION_WIDTH_THRESHOLD || element.getBBox().getX2() > page.getVisibleContentBox()
.getX2() - page.getPageWidth() * IMAGE_POSITION_WIDTH_THRESHOLD;
try (var bbox = element.getBBox(); var contentBox = page.getVisibleContentBox();) {
return bbox.getY1() < contentBox.getY1() + page.getPageHeight() * IMAGE_POSITION_HEIGHT_THRESHOLD || bbox
.getY2() >
contentBox.getY2() -
page.getPageHeight() *
IMAGE_POSITION_HEIGHT_THRESHOLD ||
bbox.getX1() < contentBox
.getX1() + page.getPageWidth() * IMAGE_POSITION_WIDTH_THRESHOLD || bbox.getX2() > contentBox
.getX2() -
page.getPageWidth() *
IMAGE_POSITION_WIDTH_THRESHOLD;
}
}
@ -214,9 +230,11 @@ public class WatermarkRemovalService {
double minAreaCoveringPage,
Page page) {
if (element.getBBox().getHeight() * element.getBBox().getWidth() < minAreaCoveringPage) {
try (var bbox = element.getBBox()) {
if (bbox.getHeight() * bbox.getWidth() < minAreaCoveringPage) {
return;
}
}
if (visitedXObjIds.add(element.getXObject().getObjNum())) {
try (ElementReader xObjectReader = new ElementReader()) {
@ -246,7 +264,9 @@ public class WatermarkRemovalService {
.filter(elementFeature -> formObjectsPerPage.values()
.stream()
.filter(elementFeaturesOnPage -> elementFeaturesOnPage.stream()
.anyMatch(elementFeature.getElementType() == Element.e_image || elementFeature.getElementType() == Element.e_inline_image ? elementFeature::isSimilarTo : elementFeature::almostMatches))
.anyMatch(
elementFeature.getElementType() == Element.e_image || elementFeature.getElementType() == Element.e_inline_image ?
elementFeature::isSimilarTo : elementFeature::almostMatches))
.count() >= minPagesFilter)
.toList();
}
@ -295,17 +315,25 @@ public class WatermarkRemovalService {
switch (element.getType()) {
case Element.e_image, Element.e_inline_image -> {
if (element.getBBox() == null) {
try (var bbox = element.getBBox()) {
if (bbox == null) {
writer.writeElement(element);
continue;
}
if (element.getBBox().getHeight() * element.getBBox().getWidth() < minAreaCoveringFromPage && isLocatedNearBorder(element, page) && element.getBBox()
.getHeight() * element.getBBox().getWidth() < minAreaCoveringFromPage || element.getXObject() == null) {
}
try (var bbox = element.getBBox()) {
if (bbox.getHeight() * bbox.getWidth() < minAreaCoveringFromPage && isLocatedNearBorder(element, page) && bbox
.getHeight() *
bbox
.getWidth() <
minAreaCoveringFromPage ||
element.getXObject() == null) {
writer.writeElement(element);
continue;
}
}
removeImages(element, writer, watermarksElementFeaturesList);
}
case Element.e_form -> processForms(page, element, reader, writer, watermarksElementFeaturesList, visitedXObjIds);
@ -344,9 +372,11 @@ public class WatermarkRemovalService {
return false;
}
if (Math.max(element.getBBox().getY1(), element.getBBox().getY2()) < page.getVisibleContentBox().getY1() + page.getPageHeight() * TEXT_POSITION_THRESHOLD) {
try (var bbox = element.getBBox(); var contents = page.getVisibleContentBox();) {
if (Math.max(bbox.getY1(), bbox.getY2()) < contents.getY1() + page.getPageHeight() * TEXT_POSITION_THRESHOLD) {
return false;
}
}
return true;
}

View File

@ -6,6 +6,7 @@ import static org.assertj.core.api.AssertionsForClassTypes.assertThat;
import java.io.FileInputStream;
import java.io.FileOutputStream;
import lombok.extern.slf4j.Slf4j;
import org.junit.jupiter.api.BeforeEach;
import org.junit.jupiter.api.Test;
@ -14,6 +15,7 @@ import com.pdftron.pdf.PDFNet;
import lombok.SneakyThrows;
@SuppressWarnings("PMD")
@Slf4j
class InvisibleElementRemovalServiceTest {
InvisibleElementRemovalService invisibleElementRemovalService;
@ -101,8 +103,19 @@ class InvisibleElementRemovalServiceTest {
String result = PdfTextExtraction.extractAllTextFromDocument(in);
assertThat(result).contains("#1 Dark",
"#13 Yellow",
"Ut wisi enim ad minim veniam, quis nostrud exerci tation ullamcorper suscipit lobortis nisl ut aliquip\n" + "ex ea commodo consequat. Duis autem vel eum iriure dolor in hendrerit in vulputate velit esse molestie\n" + "consequat, vel illum dolore eu feugiat nulla facilisis at vero eros et accumsan et iusto odio dignissim\n" + "qui blandit praesent luptatum zzril delenit augue duis dolore te feugait nulla facilisi.");
assertThat(result).doesNotContain("Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut\n" + "labore et dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et\n" + "ea rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. Lorem\n" + "ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et\n" + "dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea\n" + "rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. Lorem ipsum\n" + "dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et dolore\n" + "magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea rebum. Stet\n" + "clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. ");
"Ut wisi enim ad minim veniam, quis nostrud exerci tation ullamcorper suscipit lobortis nisl ut aliquip\n" +
"ex ea commodo consequat. Duis autem vel eum iriure dolor in hendrerit in vulputate velit esse molestie\n" +
"consequat, vel illum dolore eu feugiat nulla facilisis at vero eros et accumsan et iusto odio dignissim\n" +
"qui blandit praesent luptatum zzril delenit augue duis dolore te feugait nulla facilisi.");
assertThat(result).doesNotContain("Lorem ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut\n" +
"labore et dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et\n" +
"ea rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. Lorem\n" +
"ipsum dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et\n" +
"dolore magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea\n" +
"rebum. Stet clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. Lorem ipsum\n" +
"dolor sit amet, consetetur sadipscing elitr, sed diam nonumy eirmod tempor invidunt ut labore et dolore\n" +
"magna aliquyam erat, sed diam voluptua. At vero eos et accusam et justo duo dolores et ea rebum. Stet\n" +
"clita kasd gubergren, no sea takimata sanctus est Lorem ipsum dolor sit amet. ");
}
}