From 885fc22f9de9b4e44d7657b117b260f0a774f091 Mon Sep 17 00:00:00 2001 From: llocarnini Date: Fri, 11 Feb 2022 15:59:54 +0100 Subject: [PATCH] added changes to parse scanned pdfs --- .gitignore | 1 + vidocp/table_parsing.py | 39 +++++++++++++++++++++++++++++++++------ 2 files changed, 34 insertions(+), 6 deletions(-) create mode 100644 .gitignore diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000..bac3af5 --- /dev/null +++ b/.gitignore @@ -0,0 +1 @@ +/pdfs/ diff --git a/vidocp/table_parsing.py b/vidocp/table_parsing.py index adaa210..455e9f3 100644 --- a/vidocp/table_parsing.py +++ b/vidocp/table_parsing.py @@ -23,6 +23,19 @@ def add_external_contours(image, img): return image +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) + return img_bin + lines_h = cv2.HoughLinesP(img_bin_h, 1, np.pi/180, 500, 700, 0) + draw_lines(lines_h, img_bin_h) + + lines_v = cv2.HoughLinesP(img_bin_v, 0.7, np.pi / 180, 500, 700, 0) + draw_lines(lines_v,img_bin_v) + + return img_bin_h, img_bin_v def isolate_vertical_and_horizontal_components(img_bin): @@ -33,6 +46,18 @@ 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_final = img_bin_h | img_bin_v return img_bin_final @@ -73,15 +98,21 @@ def has_table_shape(rects): ) + + + def parse_table(image: np.array): def is_large_enough(stat): x1, y1, w, h, area = stat - # was set too higg (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 + # 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 return area > 500 and w > 35 and h > 15 gray_scale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) - #changed threshold value from 150 to 200 because of a shaded edgecase table + # 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) + + # 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) img_bin = ~img_bin @@ -96,10 +127,6 @@ 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)) - # print(f"{has_table_shape(rects) = }") - # if not has_table_shape(rects): - # print(111111111111111111111) - # return [] return rects