Even more improved line detection

main.py

@@ -12,8 +12,6 @@ for root, dirs, files in os.walk("data"):
         if file.startswith("receipt-08"):
             image = load_image("data/"+file)
             receipt = cut_receipt(image, draw_steps=True)
-            plt.imshow(receipt)
-            plt.show()
             lines = find_lines(receipt)
             for line in lines:
                 plt.imshow(line, cmap="gray")

src/processing/linefinder.py

@@ -2,15 +2,11 @@ import numpy as np
 from scipy.ndimage import measurements
 
 from src.processing.imageprocessing import rgb2gray
-from src.processing.loader import load_numpy, save_numpy, save_image, load_image
+from src.processing.loader import load_numpy, save_numpy, save_image
-from src.utils.cmap_generator import rand_cmap, list_cmap
-from matplotlib import pyplot as plt
 
 
 def find_lines(image):
     gray, binary, magnitude = preparation(image)
-    plt.imshow(binary, cmap="gray")
-    plt.show()
     backtrack = load_numpy("result/backtrack.npz")
     if backtrack is None:
         energy, backtrack = minimum_seam(binary, magnitude)
@@ -18,7 +14,8 @@ def find_lines(image):
     save_image("result/gray.png", gray)
     seams = calculate_seams(backtrack)
     labeled, ncomponents = group_empty_boxes(seams)
-    return generate_lines(labeled, ncomponents, gray)
+    lines = generate_lines(labeled, ncomponents, gray)
+    return filter_lines(lines)
 
 
 def preparation(image):
@@ -161,3 +158,14 @@ def generate_lines(labeled, ncomponents, gray):
             pixelgroup = np.concatenate((pixelgroup, pixel))
     submit_entry()
     return entries
+
+def filter_lines(lines):
+    filtered = []
+    for line in lines:
+        cnt, vals = np.histogram(line, 256)
+        threshold = get_threshold(cnt)/256*1.13#*0.96
+        binary = (line > threshold).astype(np.int_)
+        labeled, ncomponents = group_empty_boxes(binary)
+        if ncomponents > 2:
+            filtered.append(line)
+    return filtered

src/processing/receiptcutter.py

@@ -2,7 +2,6 @@ from collections import defaultdict
 
 from skimage.transform import resize
 from scipy.ndimage import gaussian_filter
-from matplotlib import pyplot as plt
 
 from src.processing.imageprocessing import rgb2gray_value
 from scipy import signal
@@ -162,8 +161,8 @@ def draw_hough_lines(image, scale, results, references, shape, theta_res=5, widt
                 y = int(n - x * m)
                 if 0 < y < image.shape[0]:
                     draw_image[max(0, y - GREEN_WIDTH):y + GREEN_WIDTH, max(0, x - GREEN_WIDTH):x + GREEN_WIDTH] = np.array([255, 0, 0])
-    plt.imshow(draw_image)
+    #plt.imshow(draw_image)
-    plt.show()
+    #plt.show()
 
 
 def convert_to_lines(scale, results, references, shape, theta_res=5, width_res=5):
@@ -343,8 +342,8 @@ def draw_rectangle(image, corners):
             x = int(a[1] + (b[1] - a[1]) * i / 5000)
             y = int(a[0] + (b[0] - a[0]) * i / 5000)
             draw_image[max(0, y - 15):y + 10, max(0, x - 10):x + 10] = np.array([255, 0, 0])
-    plt.imshow(draw_image)
+    #plt.imshow(draw_image)
-    plt.show()
+    #plt.show()
 
 
 def crop_image(image, corners):

src/utils/cmap_generator.py

@@ -2,7 +2,7 @@ from matplotlib.colors import LinearSegmentedColormap
 import colorsys
 import numpy as np
 
-def rand_cmap(nlabels, type='bright', first_color_black=True, last_color_black=False, verbose=True):
+def rand_cmap(nlabels, type='bright', first_color_black=True, last_color_black=False, verbose=False):
     """
     Creates a random colormap to be used together with matplotlib. Useful for segmentation tasks
     :param nlabels: Number of labels (size of colormap)