cv-analysis-service/cv_analysis/layout_parsing.py
Julius Unverfehrt ce9e92876c Pull request #16: Add table parsing fixtures
Merge in RR/cv-analysis from add_table_parsing_fixtures to master

Squashed commit of the following:

commit cfc89b421b61082c8e92e1971c9d0bf4490fa07e
Merge: a7ecb05 73c66a8
Author: Julius Unverfehrt <julius.unverfehrt@iqser.com>
Date:   Mon Jul 11 12:19:01 2022 +0200

    Merge branch 'master' of ssh://git.iqser.com:2222/rr/cv-analysis into add_table_parsing_fixtures

commit a7ecb05b7d8327f0c7429180f63a380b61b06bc3
Author: Julius Unverfehrt <julius.unverfehrt@iqser.com>
Date:   Mon Jul 11 12:02:07 2022 +0200

    refactor

commit 466f217e5a9ee5c54fd38c6acd28d54fc38ff9bb
Author: llocarnini <lillian.locarnini@iqser.com>
Date:   Mon Jul 11 10:24:14 2022 +0200

    deleted unused imports and unused lines of code

commit c58955c8658d0631cdd1c24c8556d399e3fd9990
Author: llocarnini <lillian.locarnini@iqser.com>
Date:   Mon Jul 11 10:16:01 2022 +0200

    black reformatted files

commit f8bcb10a00ff7f0da49b80c1609b17997411985a
Author: llocarnini <lillian.locarnini@iqser.com>
Date:   Tue Jul 5 15:15:00 2022 +0200

    reformat files

commit 432e8a569fd70bd0745ce0549c2bfd2f2e907763
Author: llocarnini <lillian.locarnini@iqser.com>
Date:   Tue Jul 5 15:08:22 2022 +0200

    added better test for generic pages with table WIP as thicker lines create inconsistent results.
    added test for patchy tables which does not work yet

commit 2aac9ebf5c76bd963f8c136fe5dd4c2d7681b469
Author: llocarnini <lillian.locarnini@iqser.com>
Date:   Mon Jul 4 16:56:29 2022 +0200

    added new fixtures for table_parsing_test.py

commit 37606cac0301b13e99be2c16d95867477f29e7c4
Author: llocarnini <lillian.locarnini@iqser.com>
Date:   Fri Jul 1 16:02:44 2022 +0200

    added separate file for table parsing fixtures, where fixtures for generic tables were added. WIP tests for generic table fixtures
2022-07-11 12:25:16 +02:00

102 lines
3.0 KiB
Python

from itertools import compress
from itertools import starmap
from operator import __and__
import cv2
import numpy as np
# from pdf2image import pdf2image
# from cv_analysis.utils.display import show_mpl
# from cv_analysis.utils.draw import draw_rectangles
from cv_analysis.utils.structures import Rectangle
from cv_analysis.utils.post_processing import (
remove_overlapping,
remove_included,
has_no_parent,
)
from cv_analysis.utils.visual_logging import vizlogger
def is_likely_segment(rect, min_area=100):
return cv2.contourArea(rect, False) > min_area
def find_segments(image):
contours, hierarchies = cv2.findContours(
image, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE
)
mask1 = map(is_likely_segment, contours)
mask2 = map(has_no_parent, hierarchies[0])
mask = starmap(__and__, zip(mask1, mask2))
contours = compress(contours, mask)
rectangles = (cv2.boundingRect(c) for c in contours)
return rectangles
def parse_layout(image: np.array):
image = image.copy()
image_ = image.copy()
if len(image_.shape) > 2:
image_ = cv2.cvtColor(image_, cv2.COLOR_BGR2GRAY)
vizlogger.debug(image_, "layout01_start.png")
image_ = cv2.GaussianBlur(image_, (7, 7), 0)
vizlogger.debug(image_, "layout02_blur.png")
thresh = cv2.threshold(image_, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
vizlogger.debug(image_, "layout03_theshold.png")
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
vizlogger.debug(kernel, "layout04_kernel.png")
dilate = cv2.dilate(thresh, kernel, iterations=4)
vizlogger.debug(dilate, "layout05_dilate.png")
rects = list(find_segments(dilate))
# -> Run meta detection on the previous detections TODO: refactor
for rect in rects:
x, y, w, h = rect
cv2.rectangle(image, (x, y), (x + w, y + h), (0, 0, 0), -1)
cv2.rectangle(image, (x, y), (x + w, y + h), (255, 255, 255), 7)
vizlogger.debug(image, "layout06_rectangles.png")
_, image = cv2.threshold(image, 254, 255, cv2.THRESH_BINARY)
vizlogger.debug(image, "layout07_threshold.png")
image = ~image
vizlogger.debug(image, "layout08_inverse.png")
if len(image.shape) > 2:
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
vizlogger.debug(image, "layout09_convertcolor.png")
rects = find_segments(image)
# <- End of meta detection
rects = remove_included(rects)
rects = remove_overlapping(rects)
return list(map(Rectangle.from_xywh, rects))
# def annotate_layout_in_pdf(page, return_rects=False, show=False):
# #page = pdf2image.convert_from_path(pdf_path, first_page=page_index + 1, last_page=page_index + 1)[0]
# #page = np.array(page)
# rects = parse_layout(page)
# if return_rects:
# return rects, page
# elif show:
# page = draw_rectangles(page, rects)
# vizlogger.debug(page, "layout10_output.png")
# show_mpl(page)
# else:
# page = draw_rectangles(page, rects)
# return page