few corrections for including smaller figures

cv_analysis/fig_detection_with_layout.py

@@ -1,41 +1,41 @@
 from cv_analysis.layout_parsing import annotate_layout_in_pdf
-from cv_analysis.figure_detection import figures_in_image, detect_figures
+from cv_analysis.figure_detection import detect_figures
-from cv_analysis.table_parsing import tables_in_image
+from cv_analysis.table_parsing import tables_in_image, parse_table
 from cv_analysis.utils.text import find_primary_text_regions, remove_primary_text_regions
 from cv_analysis.utils.draw import draw_rectangles
 from cv_analysis.utils.display import show_mpl
 
 
-def detect_parting_line(image):
-    pass
-
-
 def cut_out_content_structures(layout_rects, page):
-    large_enough_rects = []
+    large_rects = []
-    too_small_rects = []
+    small_rects = []
     for x, y, w, h in layout_rects:
         rect = (x, y, w, h)
-        if w * h >= 100000:
+        if w * h >= 50000:
             cropped_page = page[y:(y + h), x:(x + w)]
-            large_enough_rects.append([rect, cropped_page])
+            large_rects.append([rect, cropped_page])
         else:
             cropped_page = page[y:(y + h), x:(x + w)]
-            too_small_rects.append([rect, cropped_page])
+            small_rects.append([rect, cropped_page])
-    return large_enough_rects, too_small_rects
+    return large_rects, small_rects
 
 
-def parse_and_label_content_structures(page, large_enough_rects, too_small_rects):
+def parse_content_structures(page, large_rects, small_rects):
-    for coordinates, cropped_image in large_enough_rects:
+    for coordinates, cropped_image in large_rects:
         non_text_rects = detect_figures(cropped_image)
-        print(len(non_text_rects), len(list(non_text_rects)))
+
         if len(non_text_rects) == 0:
             page = draw_rectangles(page, [coordinates], color=(0, 255, 0), annotate=True)
+
         elif tables_in_image(cropped_image)[0]:
             page = draw_rectangles(page, [coordinates], color=(255, 0, 0), annotate=True)
+            stats = parse_table(page)
+            page = draw_rectangles(page, stats, annotate=True)
+
         else:
             page = draw_rectangles(page, [coordinates], color=(0, 0, 255), annotate=True)
 
-    # for coordinates, cropped_image in too_small_rects:
+    # for coordinates, cropped_image in small_rects:
     #     non_text_rects = detect_figures(cropped_image)
     #     if len(non_text_rects) == 0 and len(list(find_primary_text_regions(cropped_image))) > 0:
     #         page = draw_rectangles(page, [coordinates], color=(0, 255, 0), annotate=True)
@@ -44,16 +44,13 @@ def parse_and_label_content_structures(page, large_enough_rects, too_small_rects
     return page
 
 
-def detect_figures_over_layout():
+def detect_figures_with_layout_parsing(pdf_path, page_index=1, show=False):
-    # pdf_path = "/home/lillian/PycharmProjects/ner_address/data/pdfs/syngenta/026c917f04660aaea4bb57d180f9598b.pdf"
-    # pdf_path = "/home/lillian/ocr_docs/ocr1.pdf"
-    pdf_path = "/home/lillian/ocr_docs/Report on spectra.pdf"
-    #pdf_path = "/home/lillian/ocr_docs/VV-857853.pdf"
-    page_index = 13
     layout_rects, page = annotate_layout_in_pdf(pdf_path, page_index, return_rects=True)
     big_structures, small_structures = cut_out_content_structures(layout_rects, page)
-    page = parse_and_label_content_structures(page, big_structures, small_structures)
+    page = parse_content_structures(page, big_structures, small_structures)
+
+    if show:
         show_mpl(page)
+    else:
+        return page
 
-
-detect_figures_over_layout()

cv_analysis/figure_detection.py

@@ -1,6 +1,7 @@
 import cv2
 import numpy as np
 from pdf2image import pdf2image
+from PIL import Image
 
 from cv_analysis.utils.detection import detect_large_coherent_structures
 from cv_analysis.utils.display import show_mpl
@@ -16,9 +17,7 @@ def is_likely_figure(cont, min_area=5000, max_width_to_hight_ratio=6):
 
 def detect_figures(image: np.array):
     image = image.copy()
-    #show_mpl(image)
     image = remove_primary_text_regions(image)
-    #show_mpl(image)
     cnts = detect_large_coherent_structures(image)
 
     cnts = filter(is_likely_figure, cnts)
@@ -41,10 +40,18 @@ def detect_figures_in_pdf(pdf_path, page_index=1, show=False):
         return page
 
 
-def figures_in_image(cropped_page):
-    redaction_contours = detect_figures(cropped_page)
 
-    if len(redaction_contours) > 0:
+# pages = []
-        return True
+# for i in range(0,16):
-    else:
+#     pdf_path = "/home/lillian/ocr_docs/Report on spectra.pdf"
-        return False
+#     page_index = i
+#     page = detect_figures_in_pdf(pdf_path, page_index, show=False)
+#     pages.append(Image.fromarray(page))
+# p1, p = pages[0], pages[1:]
+#
+# out_pdf_path = "/home/lillian/ocr_docs/out.pdf"
+#
+# p1.save(
+#     out_pdf_path, "PDF", resolution=150.0, save_all=True, append_images=p
+# )
+

cv_analysis/table_parsing.py

@@ -156,10 +156,11 @@ def annotate_tables_in_pdf(pdf_path, page_index=0, deskew=False, show=False):
     else:
         return page
 
+
 def tables_in_image(cropped_image):
     table_rects = parse_table(cropped_image)
 
-    if len(table_rects)>0:
+    if len(table_rects) > 0:
         return True, table_rects
     else:
         return False, None

cv_analysis/utils/text.py

@@ -1,4 +1,6 @@
 import cv2
+import numpy as np
+
 from cv_analysis.utils.display import show_mpl
 
 def remove_primary_text_regions(image):
@@ -14,12 +16,9 @@ def remove_primary_text_regions(image):
     image = image.copy()
 
     cnts = find_primary_text_regions(image)
-
     for cnt in cnts:
         x, y, w, h = cv2.boundingRect(cnt)
-        print(x,y,w,h, w*h, w/h)
         cv2.rectangle(image, (x, y), (x + w, y + h), (255, 255, 255), -1)
-
     return image
 
 
@@ -38,7 +37,6 @@ def find_primary_text_regions(image):
 
     def is_likely_primary_text_segments(cnt):
         x,y,w,h = cv2.boundingRect(cnt)
-        print(cv2.contourArea(cnt))
         return 800 < cv2.contourArea(cnt) < 16000 or w/h > 3
 
     image = image.copy()
@@ -48,14 +46,17 @@ def find_primary_text_regions(image):
 
     image = cv2.threshold(image, 253, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
 
-    close_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (17, 7))
+    close_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (17, 7)) #20,3
     close = cv2.morphologyEx(image, cv2.MORPH_CLOSE, close_kernel, iterations=1)
-    show_mpl(close)
+
-    dilate_kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (7, 5))
+    #show_mpl(close)
+
+    dilate_kernel = cv2.getStructuringElement(cv2.MORPH_RECT,(7, 4))  #5,3
     dilate = cv2.dilate(close, dilate_kernel, iterations=1)
-    show_mpl(dilate)
+
+    #show_mpl(dilate)
+
     cnts, _ = cv2.findContours(dilate, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
     cnts = filter(is_likely_primary_text_segments, cnts)
 
     return cnts
-

scripts/annotate.py

@@ -4,13 +4,14 @@ from cv_analysis.table_parsing import annotate_tables_in_pdf
 from cv_analysis.redaction_detection import annotate_redactions_in_pdf
 from cv_analysis.layout_parsing import annotate_layout_in_pdf
 from cv_analysis.figure_detection import detect_figures_in_pdf
+from cv_analysis.fig_detection_with_layout import detect_figures_with_layout_parsing
 
 
 def parse_args():
     parser = argparse.ArgumentParser()
     parser.add_argument("pdf_path")
     parser.add_argument("page_index", type=int)
-    parser.add_argument("--type", choices=["table", "redaction", "layout", "figure"])
+    parser.add_argument("--type", choices=["table", "redaction", "layout", "figure", "figure2"])
 
     args = parser.parse_args()
 
@@ -27,3 +28,5 @@ if __name__ == "__main__":
         annotate_layout_in_pdf(args.pdf_path, page_index=args.page_index, show=True)
     elif args.type == "figure":
         detect_figures_in_pdf(args.pdf_path, page_index=args.page_index, show=True)
+    elif args.type == "figure2":
+        detect_figures_with_layout_parsing(args.pdf_path, page_index=args.page_index, show=True)