def segment_image(img):
contours = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)[1]
filtered_contours = []
for i, c in enumerate(contours):
area = cv2.contourArea(c)
if area < 1000:
filtered_contours.append(c)
filtered_contours.sort(
key=lambda c: features_util.mid_bbox(cv2.boundingRect(c))[0])
masks = []
start_x = 6
start_y = 10
width = 41
height = 70
contour_index = -1
for i in range(5):
x = start_x + (i * width)
sub_mask = img[start_y:start_y + height, x:x + width]
curr_contours = []
for j, c in enumerate(filtered_contours[contour_index:]):
mid_x, _ = features_util.mid_bbox(cv2.boundingRect(c))
if mid_x > x < x + width:
if contour_index != -1:
contour_index = j
curr_contours.append(c)
contour_index = -1
cv2.drawContours(sub_mask, curr_contours, -1, (255, 255, 255), -1)
masks.append(sub_mask)
return masks
def get_masked_images(image):
contours = cv2.findContours(image, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_TC89_L1)[0]
contours.sort(key=lambda c: features_util.mid_bbox(cv2.boundingRect(c)))
mask = np.zeros(image.shape[:2], dtype="uint8")
masks = []
curr_contours = []
sub_mask = mask.copy()
for i, c in enumerate(contours):
x, y, w, h = cv2.boundingRect(c)
mid_x = x + (w/2)
next_x = -1
if i < len(contours) - 1:
x2, y2, w2, h2 = cv2.boundingRect(contours[i+1])
next_x = x2 + (w2/2)
curr_contours.append(c)
if next_x == -1 or abs(mid_x - next_x) > 6:
cv2.drawContours(sub_mask, curr_contours, -1, (255, 255, 255), -1)
x1, y1, x2, y2 = largest_bounding_rect(curr_contours)
curr_contours = []
cropped = sub_mask[y1:y2, x1:x2]
masks.append(cropped)
sub_mask = mask.copy()
filtered_masks = []
for mask in masks:
if mask.shape[1] > 10:
filtered_masks.append(mask)
return filtered_masks
def get_masked_images(img, x_dist):
contours = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)[1]
# Sort contours by x value.
contours.sort(key=lambda c: features_util.mid_bbox(cv2.boundingRect(c))[0])
# Combine contours that are less than 'x_dist' apart from each on the x-axis (or 30 on the y).
contours = features_util.combine_contours(contours, x_dist, 40)
#contours = features_util.combine_contours_better(filtered_contours, x_dist, 30)
masks = []
for c in contours:
mask = np.zeros(img.shape, "uint8")
cv2.drawContours(mask, c, -1, (255, 255, 255), -1)
min_y, max_y, min_x, max_x = features_util.crop_to_content(mask,
padding=2)
mask = mask[min_y:max_y, min_x:max_x]
masks.append(mask)
# Split masks into two, if they are over '24' pixels in width.
# This is because some letters are too close to each other for the contours
# to be seperated.
filtered_masks = []
for mask in masks:
if mask.shape[1] > 24:
mask1 = mask[:, 0:12]
mask2 = mask[:, 12:]
filtered_masks.append(mask1)
filtered_masks.append(mask2)
elif mask.shape[0] > 12 or mask.shape[1] > 12:
filtered_masks.append(mask)
return filtered_masks
def segment_image(img):
contours = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)[0]
contours.sort(key=lambda c: features_util.mid_bbox(cv2.boundingRect(c))[0])
contours = features_util.combine_contours(contours, 3, 20)
for c in contours:
if 15 < cv2.contourArea(c[0]) < 35:
return True
return False
def get_contour_positions(img):
contours = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)[0]
contours = features_util.combine_contours(contours, 3)
bboxes = [features_util.largest_bounding_rect(c) for c in contours]
bboxes.sort(key=lambda x: x[2],
reverse=True) # Sort by width of contour bbox.
bboxes = [features_util.mid_bbox(bbox) for bbox in bboxes]
return (bboxes[0], bboxes[1])
def segment_image(img):
contours = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)[1]
min_y = 10000
min_cont = -1
for i, cont in enumerate(contours):
_, mid_y = features_util.mid_bbox(cv2.boundingRect(cont))
if mid_y < min_y:
min_y = mid_y
min_cont = i
contours.pop(min_cont)
mask = np.zeros(img.shape, dtype="uint8")
cv2.drawContours(mask, contours, -1, (255, 255, 255), 1)
cv2.drawContours(mask, contours, -1, (255, 255, 255), -1)
mask = crop_to_symbol(mask)
return mask
def segment_image(img):
contours = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)[0]
filtered_contours = []
for c in contours:
area = cv2.contourArea(c)
if area < 1000:
filtered_contours.append(c)
contours = filtered_contours
contours.sort(key=lambda c: features_util.mid_bbox(cv2.boundingRect(c))[0])
contours = features_util.combine_contours(contours, 5, 30)
masks = []
for c in contours:
mask = np.zeros(img.shape[:2], dtype="uint8")
cv2.drawContours(mask, c, -1, (255, 255, 255), -1)
masks.append(mask)
return masks