3669b6b341
layout_parsing.py added condition so fig_detection_with_layout.py works table_parsing.py uncommented line for better table parsing text.py changed kernel sizes
60 lines
2.6 KiB
Python
60 lines
2.6 KiB
Python
from cv_analysis.layout_parsing import annotate_layout_in_pdf
|
|
from cv_analysis.figure_detection import figures_in_image, detect_figures
|
|
from cv_analysis.table_parsing import tables_in_image
|
|
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 = []
|
|
too_small_rects = []
|
|
for x, y, w, h in layout_rects:
|
|
rect = (x, y, w, h)
|
|
if w * h >= 100000:
|
|
cropped_page = page[y:(y + h), x:(x + w)]
|
|
large_enough_rects.append([rect, cropped_page])
|
|
else:
|
|
cropped_page = page[y:(y + h), x:(x + w)]
|
|
too_small_rects.append([rect, cropped_page])
|
|
return large_enough_rects, too_small_rects
|
|
|
|
|
|
def parse_and_label_content_structures(page, large_enough_rects, too_small_rects):
|
|
for coordinates, cropped_image in large_enough_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)
|
|
else:
|
|
page = draw_rectangles(page, [coordinates], color=(0, 0, 255), annotate=True)
|
|
|
|
# for coordinates, cropped_image in too_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)
|
|
# else:
|
|
# page = draw_rectangles(page, [coordinates], color=(0, 255, 255), annotate=True)
|
|
return page
|
|
|
|
|
|
def detect_figures_over_layout():
|
|
# 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)
|
|
show_mpl(page)
|
|
|
|
|
|
detect_figures_over_layout()
|