format and add functions in post_processing.py missing from merge

scripts/deskew_demo.py

@@ -4,7 +4,7 @@ import numpy as np
 import pdf2image
 from PIL import Image
 
-from vidocp.utils.deskew import deskew_histbased#, deskew_linebased
+from vidocp.utils.deskew import deskew_histbased  # , deskew_linebased
 from vidocp.utils.display import show_mpl
 from vidocp.utils.draw import draw_stats
 from vidocp.table_parsing import parse_table
@@ -27,8 +27,8 @@ if __name__ == "__main__":
     page = np.array(page)
 
     show_mpl(page)
-    #page_ = deskew_linebased(page, verbose=True)
-    #show_mpl(page_)
+    # page_ = deskew_linebased(page, verbose=True)
+    # show_mpl(page_)
     page_corr = deskew_histbased(page, verbose=True)
     show_mpl(page_corr)
     if args.save_path:
@@ -36,7 +36,7 @@ if __name__ == "__main__":
         page_.save(args.save_path.replace(".pdf", "_uncorrected.pdf"))
         page_corr_ = Image.fromarray(page_corr).convert("RGB")
         page_corr_.save(args.save_path.replace(".pdf", "_corrected.pdf"))
-    #annotate_tables_in_pdf(args.pdf_path, page_index=args.page_index)
+    # annotate_tables_in_pdf(args.pdf_path, page_index=args.page_index)
     stats = parse_table(page)
     page = draw_stats(page, stats)
     show_mpl(page)
@@ -48,4 +48,3 @@ if __name__ == "__main__":
         page.save(args.save_path.replace(".pdf", "_uncorrected_annotated.pdf"))
         page_corr = Image.fromarray(page_corr).convert("RGB")
         page_corr.save(args.save_path.replace(".pdf", "_corrected_annotated.pdf"))
-    

vidocp/table_parsing.py

@@ -1,5 +1,5 @@
 from functools import partial
-from itertools import chain, starmap
+from itertools import chain, compress, starmap
 from operator import attrgetter
 
 import cv2
@@ -40,13 +40,13 @@ def isolate_vertical_and_horizontal_components(img_bin, bounding_rects):
     img_bin_v = cv2.dilate(img_bin_v, kernel_v, iterations=2)
     show_mpl(img_bin_h | img_bin_v)
 
-    #reduced filtersize from 100 to 80 to minimize splitting narrow cells
+    # reduced filtersize from 100 to 80 to minimize splitting narrow cells
     img_bin_h = apply_motion_blur(img_bin_h, 80, 0)
     img_bin_v = apply_motion_blur(img_bin_v, 80, 90)
 
     img_bin_final = img_bin_h | img_bin_v
     show_mpl(img_bin_final)
-    #changed threshold from 110 to 120 to minimize cell splitting
+    # changed threshold from 110 to 120 to minimize cell splitting
     th1, img_bin_final = cv2.threshold(img_bin_final, 120, 255, cv2.THRESH_BINARY)
     img_bin_final = cv2.dilate(img_bin_final, np.ones((1, 1), np.uint8), iterations=1)
     show_mpl(img_bin_final)
@@ -121,7 +121,7 @@ def parse_table(image: np.array):
         x1, y1, w, h, area = stat
         return area > 2000 and w > 35 and h > 25
 
-    gray_scale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) if len(image.shape)>2 else image
+    gray_scale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) if len(image.shape) > 2 else image
     th1, img_bin = cv2.threshold(gray_scale, 195, 255, cv2.THRESH_BINARY)
     img_bin = ~img_bin
     show_mpl(img_bin)

vidocp/utils/post_processing.py

@@ -1,5 +1,6 @@
 from collections import namedtuple
 from functools import partial
+from itertools import starmap, compress
 
 
 def remove_overlapping(rectangles):
@@ -19,8 +20,13 @@ def remove_included(rectangles):
     def included(a, b):
         return b.xmin >= a.xmin and b.ymin >= a.ymin and b.xmax <= a.xmax and b.ymax <= a.ymax
 
+    def includes(a, b, tol=3):
+        """does a include b?"""
+        return b.xmin + tol >= a.xmin and b.ymin + tol >= a.ymin and b.xmax - tol <= a.xmax and b.ymax - tol <= a.ymax
+
     def is_not_included(rect, rectangles):
         return not any(included(r2, rect) for r2 in rectangles if not rect == r2)
+        return not any(includes(r2, rect) for r2 in rectangles if not rect == r2)
 
     rectangles = list(map(xywh_to_vec_rect, rectangles))
     rectangles = filter(partial(is_not_included, rectangles=rectangles), rectangles)
@@ -28,19 +34,82 @@ def remove_included(rectangles):
     return rectangles
 
 
+# tolerance was set too low (1) most lines are 2px wide
+def adjacent1d(n, m, tolerance=4):
+    return abs(n - m) <= tolerance
+
+
 Rectangle = namedtuple("Rectangle", "xmin ymin xmax ymax")
 
 
+def adjacent(a, b):
+    """Two rects (v1, v2), (w1, w2) are adjacent if either of:
+    - the x components of v2 and w1 match and the y components of w1 or w2 are in the range of the y components of v1 and v2
+    - the x components of v1 and w2 match and the y components of w1 or w2 are in the range of the y components of v1 and v2
+    - the y components of v2 and w1 match and the x components of w1 or w2 are in the range of the x components of v1 and v2
+    - the y components of v1 and w2 match and the x components of w1 or w2 are in the range of the x components of v1 and v2
+    """
+
+    def adjacent2d(g, h, i, j, k, l):
+        # print(adjacent1d(g, h), any(k <= p <= l for p in [i, j]))
+        return adjacent1d(g, h) and any(k <= p <= l for p in [i, j])
+
+    if any(x is None for x in (a, b)):
+        return False
+    v1 = a.xmin, a.ymin
+    v2 = a.xmax, a.ymax
+    w1 = b.xmin, b.ymin
+    w2 = b.xmax, b.ymax
+    return any(
+        (
+            adjacent2d(v2[0], w1[0], w1[1], w2[1], v1[1], v2[1]),
+            adjacent2d(v1[0], w2[0], w1[1], w2[1], v1[1], v2[1]),
+            adjacent2d(v2[1], w1[1], w1[0], w2[0], v1[0], v2[0]),
+            adjacent2d(v1[1], w2[1], w1[0], w2[0], v1[0], v2[0]),
+        )
+    )
+
+
+# FIXME: For some reason some isolated rects remain.
+def __remove_isolated_unsorted(rectangles):
+    def is_connected(rect, rectangles):
+        return any(adjacent(r2, rect) for r2 in rectangles if not rect == r2)
+
+    rectangles = list(map(xywh_to_vec_rect, rectangles))
+    rectangles = filter(partial(is_connected, rectangles=rectangles), rectangles)
+    rectangles = map(vec_rect_to_xywh, rectangles)
+    return rectangles
+
+
 def make_box(x1, y1, x2, y2):
     keys = "x1", "y1", "x2", "y2"
     return dict(zip(keys, [x1, y1, x2, y2]))
 
 
-def compute_intersection(a, b):
+def __remove_isolated_sorted(rectangles):
+    def is_connected(left, center, right):
+        # print(left,center,right)
+        return any(starmap(adjacent, [(left, center), (center, right)]))
 
+    rectangles = list(map(xywh_to_vec_rect, rectangles))
+    lefts = [None, *rectangles[:-1]]
+    rights = [*rectangles[1:], None]
+    mask = starmap(is_connected, zip(lefts, rectangles, rights))
+    rectangles = compress(rectangles, mask)
+    rectangles = map(vec_rect_to_xywh, rectangles)
+    return rectangles
+
+
+def remove_isolated(rectangles, input_sorted=False):
+    return (__remove_isolated_sorted if input_sorted else __remove_isolated_unsorted)(rectangles)
+
+
+Rectangle = namedtuple("Rectangle", "xmin ymin xmax ymax")
+
+
+def compute_intersection(a, b):
     dx = min(a.xmax, b.xmax) - max(a.xmin, b.xmin)
     dy = min(a.ymax, b.ymax) - max(a.ymin, b.ymin)
-
     return dx * dy if (dx >= 0) and (dy >= 0) else 0
 
 
@@ -49,10 +118,21 @@ def has_no_parent(hierarchy):
 
 
 def xywh_to_vec_rect(rect):
+    v1, v2 = xywh_to_vecs(rect)
+    return Rectangle(*v1, *v2)
+
+
+def vecs_to_vec_rect(rect):
+    v1, v2 = rect
+    return Rectangle(*v1, *v2)
+
+
+def xywh_to_vecs(rect):
     x1, y1, w, h = rect
     x2 = x1 + w
     y2 = y1 + h
     return Rectangle(x1, y1, x2, y2)
+    return (x1, y1), (x2, y2)
 
 
 def vec_rect_to_xywh(rect):