From c2faf7d00bf4ed9d20d5d3422d172ae35d79643c Mon Sep 17 00:00:00 2001 From: llocarnini Date: Mon, 14 Feb 2022 11:04:04 +0100 Subject: [PATCH] adjusted isolation of vertical and horizontal components to be more robust to scanned pages; work in progress --- vidocp/table_parsing.py | 50 ++++++++++++++++++++++++----------------- 1 file changed, 30 insertions(+), 20 deletions(-) diff --git a/vidocp/table_parsing.py b/vidocp/table_parsing.py index 455e9f3..88a8790 100644 --- a/vidocp/table_parsing.py +++ b/vidocp/table_parsing.py @@ -10,6 +10,7 @@ from vidocp.utils.display import show_mpl from vidocp.utils.draw import draw_rectangles from vidocp.utils.post_processing import xywh_to_vecs, xywh_to_vec_rect, adjacent1d, remove_isolated +import matplotlib.pyplot as plt def add_external_contours(image, img): @@ -27,12 +28,12 @@ def process_lines(img_bin_h, img_bin_v): def draw_lines(lines, img_bin): for line in lines: for x1, y1, x2, y2 in line: - cv2.line(img_bin, (x1, y1), (x2, y2), (255, 255, 255), 3) + cv2.line(img_bin, (x1, y1), (x2, y2), (255, 255, 255), 6) return img_bin - lines_h = cv2.HoughLinesP(img_bin_h, 1, np.pi/180, 500, 700, 0) + lines_h = cv2.HoughLinesP(img_bin_h, 1, np.pi / 180, 500, 500, 250) draw_lines(lines_h, img_bin_h) - lines_v = cv2.HoughLinesP(img_bin_v, 0.7, np.pi / 180, 500, 700, 0) + lines_v = cv2.HoughLinesP(img_bin_v, 0.7, np.pi / 180, 500, 500, 250) draw_lines(lines_v,img_bin_v) return img_bin_h, img_bin_v @@ -46,20 +47,17 @@ def isolate_vertical_and_horizontal_components(img_bin): img_bin_h = cv2.morphologyEx(img_bin, cv2.MORPH_OPEN, kernel_h) img_bin_v = cv2.morphologyEx(img_bin, cv2.MORPH_OPEN, kernel_v) - #img_bin_h, img_bin_v = process_lines(img_bin_h,img_bin_v) - - # lines_h = cv2.HoughLinesP(img_bin_h, 1, np.pi/180, 500) - # for line in lines_h: - # for x1, y1, x2, y2 in line: - # cv2.line(img_bin_h, (x1, y1), (x2, y2), (255, 255, 255), 3) - # lines_v = cv2.HoughLinesP(img_bin_v, 0.7, np.pi / 180, 500, 600, 0) - # for line in lines_v: - # for x1, y1, x2, y2 in line: - # cv2.line(img_bin_v, (x1, y1), (x2, y2), (255, 255, 255), 3) - + img_bin_h = cv2.dilate(img_bin_h, kernel_h, 1) + img_bin_v = cv2.dilate(img_bin_v, kernel_v, 1) + img_bin_h = apply_motion_blur(img_bin_h, 100, 0) + img_bin_v = apply_motion_blur(img_bin_v, 100, 90) + # img_bin_h, img_bin_v = process_lines(img_bin_h,img_bin_v) img_bin_final = img_bin_h | img_bin_v - + kernel = np.ones((5, 5), np.uint8) + # img_bin_final = cv2.dilate(img_bin_final, kernel, 2) + th1, img_bin_final = cv2.threshold(img_bin_final, 10, 255, cv2.THRESH_BINARY) + show_mpl(img_bin_final) return img_bin_final @@ -99,21 +97,27 @@ def has_table_shape(rects): +def apply_motion_blur(image, size, angle): + k = np.zeros((size, size), dtype=np.float32) + k[ (size-1)// 2 , :] = np.ones(size, dtype=np.float32) + k = cv2.warpAffine(k, cv2.getRotationMatrix2D( (size / 2 -0.5 , size / 2 -0.5 ) , angle, 1.0), (size, size) ) + k = k * ( 1.0 / np.sum(k) ) + return cv2.filter2D(image, -1, k) def parse_table(image: np.array): def is_large_enough(stat): x1, y1, w, h, area = stat # was set too high (3000): Boxes in a Table can be smaller. example: a column titled "No." This cell has approximatly an area of 500 px based on 11pt letters - # with extra condition for the length of height and width, weirdly narrow rectangles can be filtered + # with extra condition for the length of height and width weirdly narrow rectangles can be filtered return area > 500 and w > 35 and h > 15 gray_scale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) - # blur_gray_scale = cv2.GaussianBlur(gray_scale, (5, 5), 1, borderType=cv2.BORDER_REPLICATE) - # th1, img_bin = cv2.threshold(blur_gray_scale, 195, 255, cv2.THRESH_BINARY) - + blur_gray_scale = cv2.GaussianBlur(gray_scale, (5, 5), 1, borderType=cv2.BORDER_REPLICATE) + th1, img_bin = cv2.threshold(blur_gray_scale, 195, 255, cv2.THRESH_BINARY) + show_mpl(img_bin) # changed threshold value from 150 to 195 because of a shaded edgecase table - th1, img_bin = cv2.threshold(gray_scale, 195, 255, cv2.THRESH_BINARY) + # th1, img_bin = cv2.threshold(gray_scale, 195, 255, cv2.THRESH_BINARY) img_bin = ~img_bin img_bin = isolate_vertical_and_horizontal_components(img_bin) @@ -127,6 +131,10 @@ def parse_table(image: np.array): # FIXME: produces false negatives for `data0/043d551b4c4c768b899eaece4466c836.pdf 1 --type table` rects = list(remove_isolated(rects, input_sorted=True)) + # if not has_table_shape(rects): + # return False + + return rects @@ -138,5 +146,7 @@ def annotate_tables_in_pdf(pdf_path, page_index=1): stats = parse_table(page) page = draw_rectangles(page, stats, annotate=True) + # if stats: + # page = draw_rectangles(page, stats, annotate=True) show_mpl(page)