103 lines
2.8 KiB
Python
103 lines
2.8 KiB
Python
from functools import reduce
|
|
from itertools import compress
|
|
from operator import __and__
|
|
from typing import Iterable
|
|
|
|
import cv2
|
|
import numpy as np
|
|
from funcy import lmap, compose
|
|
|
|
from cv_analysis.utils.connect_rects import connect_related_rects2
|
|
from cv_analysis.utils.conversion import box_to_rectangle, rectangle_to_box
|
|
from cv_analysis.utils.postprocessing import (
|
|
remove_included,
|
|
has_no_parent,
|
|
)
|
|
from cv_analysis.utils.rectangle import Rectangle
|
|
|
|
|
|
def parse_layout(image: np.array):
|
|
|
|
rectangles = find_segments(image)
|
|
|
|
rectangles = remove_included(rectangles)
|
|
rectangles = lmap(rectangle_to_box, rectangles)
|
|
rectangles = connect_related_rects2(rectangles)
|
|
rectangles = lmap(box_to_rectangle, rectangles)
|
|
rectangles = remove_included(rectangles)
|
|
|
|
return rectangles
|
|
|
|
|
|
def find_segments(image, meta=1):
|
|
|
|
if meta:
|
|
original = image.copy()
|
|
image = normalize_to_gray_scale(image)
|
|
image = dilate_page_components(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 = map(__and__, mask1, mask2)
|
|
contours = compress(contours, mask)
|
|
rectangles = lmap(compose(box_to_rectangle, cv2.boundingRect), contours)
|
|
|
|
if meta:
|
|
image = meta_detection(original, rectangles)
|
|
rectangles = find_segments(image, 0)
|
|
|
|
return rectangles
|
|
|
|
|
|
def is_likely_segment(rect, min_area=100):
|
|
return cv2.contourArea(rect, False) > min_area
|
|
|
|
|
|
def dilate_page_components(image):
|
|
image = cv2.GaussianBlur(image, (7, 7), 0)
|
|
thresh = cv2.threshold(image, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)[1]
|
|
kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
|
|
return cv2.dilate(thresh, kernel, iterations=4)
|
|
|
|
|
|
def meta_detection(image: np.ndarray, rectangles: Iterable[Rectangle]):
|
|
"""Given a list of previously detected segments, rerun the detection algorithm. Heuristically this improves the
|
|
quality of the detection.
|
|
"""
|
|
|
|
image = fill_rectangles(image, rectangles)
|
|
image = threshold_image(image)
|
|
image = invert_image(image)
|
|
image = normalize_to_gray_scale(image)
|
|
|
|
return image
|
|
|
|
|
|
def normalize_to_gray_scale(image):
|
|
if len(image.shape) > 2:
|
|
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
|
|
return image
|
|
|
|
|
|
def threshold_image(image):
|
|
_, image = cv2.threshold(image, 254, 255, cv2.THRESH_BINARY)
|
|
return image
|
|
|
|
|
|
def invert_image(image):
|
|
return ~image
|
|
|
|
|
|
def fill_rectangles(image, rectangles):
|
|
image = reduce(fill_in_component_area, rectangles, image)
|
|
return image
|
|
|
|
|
|
def fill_in_component_area(image, rect):
|
|
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)
|
|
|
|
return image
|