change default deskew function from hough-line-based to pixel-histogram-based; use scipy.ndimage.rotation

requirements.txt

@@ -6,4 +6,4 @@ imutils==0.5.4
 iteration-utilities==0.11.0
 dvc==2.9.3
 dvc[ssh]
-
+scipy~=1.8.0

scripts/deskew_demo.py

@@ -4,7 +4,7 @@ import numpy as np
 import pdf2image
 
 from vidocp.utils.display import show_mpl
-from vidocp.utils.deskew import deskew_image
+from vidocp.utils.deskew import deskew_histbased, deskew_linebased
 
 
 def parse_args():
@@ -23,5 +23,7 @@ if __name__ == "__main__":
     page = np.array(page)
 
     show_mpl(page)
-    page = deskew_image(page, verbose=True)
+    page_ = deskew_linebased(page, verbose=True)
-    show_mpl(page)
+    show_mpl(page_)
+    page_ = deskew_histbased(page, verbose=True)
+    show_mpl(page_)

vidocp/table_parsing.py

@@ -4,6 +4,7 @@ from pdf2image import pdf2image
 
 from vidocp.utils.display import show_mpl
 from vidocp.utils.draw import draw_stats
+from vidocp.utils.deskew import deskew_histbased
 
 
 def add_external_contours(image, img):
@@ -33,7 +34,7 @@ def isolate_vertical_and_horizontal_components(img_bin):
 
 def parse_table(image: np.array):
 
-    gray_scale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
+    gray_scale = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) if len(image.shape)>2 else image
     th1, img_bin = cv2.threshold(gray_scale, 150, 255, cv2.THRESH_BINARY)
     img_bin = ~img_bin
 
@@ -45,10 +46,12 @@ def parse_table(image: np.array):
     return stats
 
 
-def annotate_tables_in_pdf(pdf_path, page_index=1):
+def annotate_tables_in_pdf(pdf_path, page_index=1, deskew=True):
 
     page = pdf2image.convert_from_path(pdf_path, first_page=page_index + 1, last_page=page_index + 1)[0]
     page = np.array(page)
+    if deskew:
+        page = deskew_histbased(page)
 
     stats = parse_table(page)
     page = draw_stats(page, stats)

vidocp/utils/deskew.py

@@ -1,8 +1,9 @@
 import numpy as np
+from scipy.ndimage import rotate
 import cv2
 
 
-def detect_angle(im: np.array, max_skew_deg=10, min_skew_deg=0.1, min_nlines=5) -> int:
+def detect_angle_from_lines(im: np.array, max_skew_deg=10, min_skew_deg=0.1, min_nlines=5) -> int:
     max_skew_rad = np.deg2rad(max_skew_deg)
     min_skew_rad = np.deg2rad(min_skew_deg)
     width = im.shape[1]
@@ -41,11 +42,35 @@ def rotate_straight(im: np.array, skew_angle: int) -> np.array:
     return rotated
 
 
-def deskew_image(image: np.array, verbose=False) -> np.array:
+def deskew_linebased(image: np.array, verbose=False) -> np.array:
-    skew_angle = detect_angle(image)
+    skew_angle = detect_angle_from_lines(image)
     if verbose:
         print(f"Skew angle: {skew_angle}")
     if skew_angle:
         deskewed = rotate_straight(image, skew_angle)
         return deskewed
     return image
+
+
+def deskew_histbased(page: np.array, max_abs_angle=1.5, delta=0.15, mode="nearest", verbose=False):
+    page = cv2.cvtColor(page, cv2.COLOR_BGR2GRAY)
+    page = cv2.fastNlMeansDenoising(page, h=3)
+    w, h = page.shape
+
+    def find_score(arr, angle):
+        data = rotate(arr, angle, reshape=False, order=0)
+        hist = np.sum(data, axis=1)
+        score = np.sum((hist[1:] - hist[:-1]) ** 2)
+        return score
+
+    angles = np.arange(-max_abs_angle, max_abs_angle + delta, delta)
+    scores = []
+    for angle in angles:
+        scores.append(find_score(page, angle))
+
+    best_angle = angles[scores.index(max(scores))]
+    if verbose:
+        print("Best angle: {}".format(best_angle))
+
+    rotated = rotate(page, best_angle, reshape=False, order=0, mode=mode)
+    return rotated

vidocp/utils/display.py

@@ -3,14 +3,12 @@ from matplotlib import pyplot as plt
 
 
 def show_mpl(image):
-
     fig, ax = plt.subplots(1, 1)
     fig.set_size_inches(20, 20)
-    ax.imshow(image)
+    ax.imshow(image, cmap="gray")
     plt.show()
 
 
 def show_cv2(image):
-
     cv2.imshow("", image)
     cv2.waitKey(0)