def smoothBtn(self): """ Smooth button clicked! :return: """ print("Smooth button clicked") if self.scene.points is None or len(self.scene.points) == 0: return # max Error max_error = int(self.maxerror_ledit.text()) # new contour image contour_img = np.array(np.ones_like(self.contour_gray) * 255, dtype=np.uint8) # smooth the contour segmentations. contour_sorted = sortPointsOnContourOfImage(self.contour_gray.copy()) feature_points = [] for pt in self.scene.points: nearest_pt = None max_dist = 1000000 for cpt in contour_sorted: dist_ = math.sqrt((pt[0] - cpt[0])**2 + (pt[1] - cpt[1])**2) if dist_ < max_dist: max_dist = dist_ nearest_pt = cpt # select feature points feature_points.append(nearest_pt) # add first point as the last end point feature_points.append(feature_points[0]) print(self.scene.points) print(feature_points) self.feature_points = feature_points.copy() # extract segmentations based on the feature points contour_segmentations = [] for id in range(len(feature_points) - 1): start_pt = feature_points[id] end_pt = feature_points[id + 1] start_index = contour_sorted.index(start_pt) end_index = contour_sorted.index(end_pt) if start_index < end_index: segmentation = contour_sorted[start_index:end_index] if end_index == len(contour_sorted) - 1: segmentation.append(contour_sorted[0]) else: segmentation.append(contour_sorted[end_index + 1]) elif start_index >= end_index: segmentation = contour_sorted[start_index:len( contour_sorted)] + contour_sorted[0:end_index + 1] contour_segmentations.append(segmentation) print("contour segmentation len: %d" % len(contour_segmentations)) self.contour_segmentations = contour_segmentations.copy() # smooth contour segmentations smoothed_contour_points = [] for id in range(len(self.contour_segmentations)): print("Line index: %d" % id) # smooth contour segmentation li_seg = np.array(self.contour_segmentations[id]) beziers = fitCurve(li_seg, maxError=max_error) for bez in beziers: bezier_points = draw_cubic_bezier(bez[0], bez[1], bez[2], bez[3]) for id in range(len(bezier_points) - 1): start_pt = bezier_points[id] end_pt = bezier_points[id + 1] cv2.line(contour_img, start_pt, end_pt, color=0, thickness=1) smoothed_contour_points += bezier_points qimg = QImage(contour_img.data, contour_img.shape[1], contour_img.shape[0], contour_img.shape[1], \ QImage.Format_Indexed8) contour_pix = QPixmap.fromImage(qimg) self.scene.addPixmap(contour_pix) self.scene.update() # clean data self.scene.points = [] self.contour_gray = contour_img.copy() self.smoothed_contour_points = smoothed_contour_points.copy() # update status bar self.statusbar.showMessage("Smooth successed!") del contour_img, contour_pix, contour_segmentations, contour_sorted, smoothed_contour_points, \ qimg, feature_points
def autoSmoothContoursOfComponent(component, blockSize=3, ksize=3, k=0.04): """ :param component: :return: """ if component is None: return # 5. Using corner detection to get corner regions corner_component = np.float32(component) dst = cv2.cornerHarris(corner_component, blockSize=blockSize, ksize=ksize, k=k) dst = cv2.dilate(dst, None) corners_area_points = [] for y in range(dst.shape[0]): for x in range(dst.shape[1]): if dst[y][x] > 0.1 * dst.max(): corners_area_points.append((x, y)) print("corner area points num: %d" % len(corners_area_points)) # 6. Determine center points of corner areas blank_gray = createBlankGrayscaleImage(component) for pt in corners_area_points: blank_gray[pt[1]][pt[0]] = 0.0 rectangles = getAllMiniBoundingBoxesOfImage(blank_gray) corners_area_center_points = [] for rect in rectangles: corners_area_center_points.append( (rect[0] + int(rect[2] / 2.), rect[1] + int(rect[3] / 2.))) print("corner area center points num: %d" % len(corners_area_center_points)) # based the distance to end points and cross points, remove extra corners area center points component_skeleton = getSkeletonOfImage(component) end_points = getEndPointsOfSkeletonLine(component_skeleton) cross_points = getCrossPointsOfSkeletonLine(component_skeleton) # remove extra branches # img_skeleton = removeBranchOfSkeletonLine(img_skeleton, end_points, cross_points) # end_points = getEndPointsOfSkeletonLine(img_skeleton) # cross_points = getEndPointsOfSkeletonLine(img_skeleton) # detect valid corner region center points closed to end points and cross points valid_corners_area_center_points = [] dist_threshold = 40 for pt in corners_area_center_points: is_valid = False for ept in end_points: dist = math.sqrt((pt[0] - ept[0])**2 + (pt[1] + ept[1])**2) if dist <= dist_threshold: is_valid = True break if is_valid: valid_corners_area_center_points.append(pt) continue for cpt in cross_points: dist = math.sqrt((pt[0] - cpt[0])**2 + (pt[1] - cpt[1])**2) if dist <= dist_threshold: is_valid = True break if is_valid: valid_corners_area_center_points.append(pt) print("valid corner area center points num: %d" % len(valid_corners_area_center_points)) del blank_gray # 7. Get all contours of component component_contours = getContourOfImage(component) contours = getConnectedComponents(component_contours, connectivity=8) print("contours num: %d" % len(contours)) # 8. Process contours to get closed and 1-pixel width contours contours_processed = [] for cont in contours: cont = removeBreakPointsOfContour(cont) contours_processed.append(cont) print("contours processed num: %d" % len(contours_processed)) # 9. Find corner points of conthours closed to corner region center points. For each contour, there is a coner points list. contours_corner_points = [] for i in range(len(contours_processed)): corner_points = [] contour = contours_processed[i] for pt in valid_corners_area_center_points: x0 = target_x = pt[0] y0 = target_y = pt[1] min_dist = 10000 # search target point in region: 20 * 20 of center is (x0, y0) for y in range(y0 - 10, y0 + 10): for x in range(x0 - 10, x0 + 10): if contour[y][x] == 255: continue dist = math.sqrt((x - x0)**2 + (y - y0)**2) if dist < min_dist: min_dist = dist target_x = x target_y = y if min_dist < 5: corner_points.append((target_x, target_y)) contours_corner_points.append(corner_points) total_num = 0 for cont in contours_corner_points: total_num += len(cont) if total_num == len(valid_corners_area_center_points): print("corner points not ignored") else: print("corner points be ignored") # 10. Separate contours into sub-contours based on the corner points on different contours sub_contours = [] for i in range(len(contours_processed)): contour = contours_processed[i] corner_points = contours_corner_points[i] # sorted the contour contour_points_sorted = sortPointsOnContourOfImage(contour) # sorted the corner points corner_points_sorted = [] for pt in contour_points_sorted: if pt in corner_points: corner_points_sorted.append(pt) # sepate the contour into sub-contour for j in range(len(corner_points_sorted)): start_pt = corner_points_sorted[j] end_pt = None if j == len(corner_points_sorted) - 1: end_pt = corner_points_sorted[0] else: end_pt = corner_points_sorted[j + 1] # find indexes of start point and end point in contour_points_sorted start_index = contour_points_sorted.index(start_pt) end_index = contour_points_sorted.index(end_pt) # separate sub_contour = None if start_index <= end_index: if end_index == len(contour_points_sorted) - 1: sub_contour = contour_points_sorted[ start_index:len(contour_points_sorted)] sub_contour.append(contour_points_sorted[0]) else: sub_contour = contour_points_sorted[start_index:end_index + 1] else: sub_contour = contour_points_sorted[ start_index:len(contour_points_sorted )] + contour_points_sorted[0:end_index + 1] sub_contours.append(sub_contour) print("sub contours num: %d" % len(sub_contours)) # 11. Beizer curve fit all sub-contours under maximal error max_error = 100 sub_contours_smoothed = [] for id in range(len(sub_contours)): # single sub-contour sub_contour = np.array(sub_contours[id]) if len(sub_contour) < 2: continue beziers = fitCurve(sub_contour, maxError=max_error) sub_contour_smoothed = [] for bez in beziers: bezier_points = draw_cubic_bezier(bez[0], bez[1], bez[2], bez[3]) sub_contour_smoothed += bezier_points sub_contours_smoothed.append(sub_contour_smoothed) # 12. Merge sub-contours together img_smoothed_gray = createBlankGrayscaleImage(component) # merge all smoothed sub-contours for sub in sub_contours_smoothed: for pt in sub: img_smoothed_gray[pt[1]][pt[0]] = 0.0 # process smoothed contours to get closed and 1-pixel width img_smoothed_gray = getSkeletonOfImage(img_smoothed_gray) # remove single points that 8 cv2.imshow("img_smoothed_gray", img_smoothed_gray) contours_smoothed = getConnectedComponents(img_smoothed_gray) if len(contours_smoothed) == 1: # no hole exist, directly fill black in the contour cont = contours_smoothed[0] cont_points = sortPointsOnContourOfImage(cont) cont_points = np.array([cont_points], "int32") fill_contour_smooth = np.ones_like(component) * 255 fill_contour_smooth = np.array(fill_contour_smooth, dtype=np.uint8) fill_contour_smooth = cv2.fillPoly(fill_contour_smooth, cont_points, 0) return fill_contour_smooth else: # exist hole, should processed print("there are holes!") fill_img_list = [] hole_points = [] for cont in contours_smoothed: cont_points = sortPointsOnContourOfImage(cont) cont_points = np.array([cont_points], "int32") fill_contour_smooth = np.ones_like(component) * 255 fill_contour_smooth = np.array(fill_contour_smooth, dtype=np.uint8) fill_contour_smooth = cv2.fillPoly(fill_contour_smooth, cont_points, 0) valid_num = same_num = 0 for y in range(component.shape[0]): for x in range(component.shape[1]): if component[y][x] == 0.0: valid_num += 1 if fill_contour_smooth[y][x] == 0.0: same_num += 1 if 1.0 * same_num / valid_num > 0.8: fill_img_list.append(fill_contour_smooth) print("ratio: %f" % (1.0 * same_num / valid_num)) else: print("ratio: %f" % (1.0 * same_num / valid_num)) for y in range(fill_contour_smooth.shape[0]): for x in range(fill_contour_smooth.shape[1]): if fill_contour_smooth[y][x] == 0.0: hole_points.append((x, y)) # merge filled images blank_temp = np.ones_like(component) * 255 for fl in fill_img_list: for y in range(fl.shape[0]): for x in range(fl.shape[1]): if fl[y][x] == 0.0: blank_temp[y][x] = fl[y][x] # hole points for pt in hole_points: blank_temp[pt[1]][pt[0]] = 255 return blank_temp
def autoStrokeExtractFromComponent(component): """ Automatically strokes extract from the component. :param component: :return: """ strokes = [] if component is None: return strokes # 6. Get skeletons of component. comp_skeleton = getSkeletonOfImage(component.copy()) # cv2.imshow("skeleton_original", comp_skeleton) # 7. Process the skeleton by remove extra branches. comp_skeleton = removeShortBranchesOfSkeleton(comp_skeleton, length_threshold=30) # cv2.imshow("skeleton_smoothed", comp_skeleton) # 8. Get the end points and cross points after skeleton processed end_points = getEndPointsOfSkeletonLine(comp_skeleton) cross_points = getCrossPointsOfSkeletonLine(comp_skeleton) print("end points num: %d ,and cross points num: %d" % (len(end_points), len(cross_points))) # 9. Get contour image of component comp_contours_img = getContourImage(component.copy()) # 10. Detect the number of contours and return all contours comp_contours = getConnectedComponents(comp_contours_img, connectivity=8) print("contours num: %d" % len(comp_contours)) # 11. Get points on contours corners_points = [] for cont in comp_contours: cont = removeBreakPointsOfContour(cont) cont_sorted = sortPointsOnContourOfImage(cont) cont_points = rdp(cont_sorted, 5) corners_points += cont_points print("corner points num:", len(corners_points)) CORNER_CROSS_DIST_THRESHOLD = 30 corners_points_merged = [] for pt in corners_points: for cpt in cross_points: dist = math.sqrt((pt[0] - cpt[0]) ** 2 + (pt[1] - cpt[1]) ** 2) if dist < CORNER_CROSS_DIST_THRESHOLD: corners_points_merged.append(pt) break corners_points = corners_points_merged print("merged corner points num:", len(corners_points)) # # 11. Detect the corner regions of component # # Harris corner detector # corner_region_img = np.float32(component.copy()) # dst = cv2.cornerHarris(corner_region_img, blockSize=3, ksize=3, k=0.04) # dst = cv2.dilate(dst, None) # # # get all points in corners area # corners_area_points = [] # for y in range(dst.shape[0]): # for x in range(dst.shape[1]): # if dst[y][x] > 0.1 * dst.max(): # corners_area_points.append((x, y)) # # # get all center points of corner area # corners_img = createBlankGrayscaleImage(component) # for pt in corners_area_points: # corners_img[pt[1]][pt[0]] = 0.0 # # rectangles = getAllMiniBoundingBoxesOfImage(corners_img) # # corners_area_center_points = [] # for rect in rectangles: # corners_area_center_points.append((rect[0] + int(rect[2] / 2.), rect[1] + int(rect[3] / 2.))) # # # get all corner points in coutour image. # corners_points = [] # for pt in corners_area_center_points: # if comp_contours_img[pt[1]][pt[0]] == 0.0: # corners_points.append(pt) # else: # min_dist = 100000 # min_x = min_y = 0 # for y in range(comp_contours_img.shape[0]): # for x in range(comp_contours_img.shape[1]): # cpt = comp_contours_img[y][x] # if cpt == 0.0: # dist = math.sqrt((x - pt[0]) ** 2 + (y - pt[1]) ** 2) # if dist < min_dist: # min_dist = dist # min_x = x # min_y = y # # points on contour # corners_points.append((min_x, min_y)) print("corners points num: %d" % len(corners_points)) # 12. Get valid corner points based on the end points and cross points corners_points = getValidCornersPoints(corners_points, cross_points, end_points, distance_threshold=30) print("corners points num: %d" % len(corners_points)) if len(corners_points) == 0: print("no corner points") strokes.append(component) return strokes # 13. Cluster these corner points based on the distance between them and cross points corners_points_cluster = getClusterOfCornerPoints(corners_points, cross_points, threshold_distance=70) print("corner points cluster num: %d" % len(corners_points_cluster)) print(corners_points_cluster) # 14. Generate cropping lines between two corner points crop_lines = getCropLines(corners_points_cluster, comp_contours) print("cropping lines num: %d" % len(crop_lines)) # 15. Separate the components based on the cropping lines component_ = component.copy() # add white and 1-pixel width line in component to separate it. for line in crop_lines: cv2.line(component_, line[0], line[1], 255, 1) # 16. Get parts of component. comp_parts = [] # invert color !!! component_ = 255 - component_ ret, labels = cv2.connectedComponents(component_, connectivity=4) print(ret) for r in range(1, ret): img_ = createBlankGrayscaleImage(component_) for y in range(component_.shape[0]): for x in range(component_.shape[1]): if labels[y][x] == r: img_[y][x] = 0.0 if img_[0][0] != 0.0 and isValidComponent(img_, component): comp_parts.append(img_) print("parts of component num: %d" % len(comp_parts)) # 17. Add cropping lines to corresponding parts of component # add lines to parts of component. for i in range(len(comp_parts)): part = comp_parts[i] for line in crop_lines: start_dist = getDistanceBetweenPointAndComponent(line[0], part) end_dist = getDistanceBetweenPointAndComponent(line[1], part) if start_dist <= 3 and end_dist <= 3: cv2.line(part, line[0], line[1], 0, 1) # 18. Find intersection parts of component used_index = [] intersect_parts_index = [] for i in range(len(comp_parts)): part = comp_parts[i] num = 0 # number of lines in part for line in crop_lines: if part[line[0][1]][line[0][0]] == 0.0 and part[line[1][1]][line[1][0]] == 0.0: num += 1 if num == 4: # 4 lines in one part, this part is the intersection part intersect_parts_index.append(i) used_index.append(i) print("intersection parts num: %d" % len(intersect_parts_index)) # 19. Find the relation part and crop lines - one line -> one part or three part (part1 + intersect_part + part2) intersect_parts_crop_lines_index = [] for i in range(len(crop_lines)): line = crop_lines[i] for index in intersect_parts_index: intersect_part = comp_parts[index] if intersect_part[line[0][1]][line[0][0]] == 0.0 and intersect_part[line[1][1]][line[1][0]] == 0.0: # this line in intersection part intersect_parts_crop_lines_index.append(i) print("crop lines in intersection part num: %d" % len(intersect_parts_crop_lines_index)) # Cropping lines are divided into two types: in intersect part and not in this part. line_parts_relation = [] for index in intersect_parts_crop_lines_index: line = crop_lines[index] # line and parts that are connected by this crop line: A - intersect_part - B line_connected_parts = [] # find intersection part contains this line for i in intersect_parts_index: intersect_part = comp_parts[i] if intersect_part[line[0][1]][line[0][0]] == 0.0 and intersect_part[line[1][1]][line[1][0]] == 0.0: # line in this intersect part line_connected_parts.append(i) # find two parts connectd by this crop line for i in range(len(comp_parts)): # part should not be the intersect part if i in intersect_parts_index: continue # check only end point of line in part. part = comp_parts[i] if part[line[0][1]][line[0][0]] == 0.0 and part[line[1][1]][line[1][0]] != 0.0 or \ part[line[0][1]][line[0][0]] != 0.0 and part[line[1][1]][line[1][0]] == 0.0: line_connected_parts.append(i) # add line connected parts to relation list. if line_connected_parts not in line_parts_relation: line_parts_relation.append(line_connected_parts) # add independent parts to relation of line and parts for i in range(len(comp_parts)): line_connected_parts = [] is_independent = True for relation in line_parts_relation: if i in relation: is_independent = False # check this part is independent or not if is_independent: line_connected_parts.append(i) if line_connected_parts != []: line_parts_relation.append(line_connected_parts) # 20. Merge parts based on the line parts relation for i in range(len(line_parts_relation)): # blank image blank_ = createBlankGrayscaleImage(component) # parts relation list relation = line_parts_relation[i] for rel in relation: part = comp_parts[rel] if part is None: continue # merge part and blank image for y in range(part.shape[0]): for x in range(part.shape[1]): if part[y][x] == 0.0: blank_[y][x] = 0.0 # add to strokes list strokes.append(blank_) return strokes
def autoStrokeExtracting(index, image, threshold_value=200): """ Automatic strokes extracting :param image: grayscale image :return: strokes images with same size """ strokes = [] if image is None: return strokes # get connected components from the grayscale image, not for the binary image. contour_img = getContourImage(image) contours = getConnectedComponents(contour_img) # no holes, num=1, holes exist, num >= 2 print("contours num: %d" % len(contours)) corners_points_sorted = [] for ct in contours: points = sortPointsOnContourOfImage(ct) corners_points_sorted.append(points) if len(corners_points_sorted) == 1: print("No holes exist!") elif len(corners_points_sorted) >= 2: print("Holes exist!") # grayscale image to binary image _, img_bit = cv2.threshold(image, threshold_value, 255, cv2.THRESH_BINARY) # skeleton image of width 1 pixel of binray image skeleton_img = getSkeletonOfImage(img_bit) skeleton_img = removeExtraBranchesOfSkeleton(skeleton_img) end_points = getEndPointsOfSkeletonLine(skeleton_img) # end points cross_points = getCrossPointsOfSkeletonLine(skeleton_img) # croiss points print("end points num: %d" % len(end_points)) print("cross points num: %d" % len(cross_points)) if len(cross_points) == 0: print("no cross points!") strokes.append(image) return strokes # corner area points corners_all_points = getCornersPoints(image.copy(), contour_img, blockSize=3, ksize=3, k=0.04) corners_points = getValidCornersPoints(corners_all_points, cross_points, end_points, distance_threshold=30) print("corners points num: %d" % len(corners_points)) if len(corners_points) == 0: print("no corner point") strokes.append(image) return strokes contour_rgb = cv2.cvtColor(contour_img, cv2.COLOR_GRAY2RGB) contour_gray = cv2.cvtColor(contour_rgb, cv2.COLOR_RGB2GRAY) _, contour_gray = cv2.threshold(contour_gray, 200, 255, cv2.THRESH_BINARY) for pt in corners_points: contour_rgb[pt[1]][pt[0]] = (0, 0, 255) # cluster corners points based on the cross point corner_points_cluster = getClusterOfCornerPoints(corners_points, cross_points) # cropping lines based on the corner points crop_lines = getCropLines(corner_points_cluster, None) for line in crop_lines: cv2.line(contour_rgb, line[0], line[1], (0, 255, 0), 1) cv2.line(contour_gray, line[0], line[1], 0, 1) # split contour to components ret, labels = cv2.connectedComponents(contour_gray, connectivity=4) components = [] for r in range(1, ret): img_ = createBlankGrayscaleImage(contour_gray) for y in range(contour_gray.shape[0]): for x in range(contour_gray.shape[1]): if labels[y][x] == r: img_[y][x] = 0.0 if img_[0][0] != 0.0 and isValidComponent(img_, img_bit): components.append(img_) print("components num : %d" % len(components)) used_components = [] component_line_relation = {} # {component_id: [line_id1, line_id2]} # merge contour to components for i in range(len(components)): merge_points = [] for y in range(1, contour_gray.shape[0]-1): for x in range(1, contour_gray.shape[1]-1): if contour_gray[y][x] == 0.0: # 4 nearby pixels should be black in components if components[i][y-1][x] == 0.0 or components[i][y][x+1] == 0.0 or components[i][y+1][x] == 0.0 or \ components[i][y][x-1] == 0.0: merge_points.append((x, y)) for pt in merge_points: components[i][pt[1]][pt[0]] = 0.0 # merge cropping lines on the components for i in range(len(crop_lines)): components_id = [] line = crop_lines[i] for j in range(len(components)): dist_startpt = getDistanceBetweenPointAndComponent(line[0], components[j]) # print("dist startpt:%f" % dist_startpt) dist_endpt = getDistanceBetweenPointAndComponent(line[1], components[j]) # print("dist end pt: %f" % dist_endpt) if dist_startpt < 3 and dist_endpt < 3: cv2.line(components[j], line[0], line[1], 0, 1) components_id.append(j) if len(components_id) >= 2: break # find overlap region components overlap_components = [] for i in range(len(components)): part = components[i] part_lines = [] part_lines_id = [] for j in range(len(crop_lines)): line = crop_lines[j] if part[line[0][1]][line[0][0]] == 0.0 and part[line[1][1]][line[1][0]] == 0.0: part_lines.append(line) part_lines_id.append(j) # check number of lines == 4 and cross each other if len(part_lines) == 4: points_set = set() for line in part_lines: points_set.add(line[0]) points_set.add(line[1]) if len(points_set) == 4: # only 4 points overlap_components.append(part) used_components.append(i) component_line_relation[i] = part_lines_id print("overlap components num: %d" % len(overlap_components)) print(component_line_relation) # cluster components based on the cropping lines for i in range(len(components)): part = components[i] part_lines = [] # used to detect overlap region components. # find single part is stroke is_single = True for line in crop_lines: x1 = line[0][0]; y1 = line[0][1] x2 = line[1][0]; y2 = line[1][1] if part[y1][x1] == 0.0 and part[y2][x2] != 0.0 or part[y1][x1] != 0.0 and part[y2][x2] == 0.0: is_single = False break if part[y1][x1] == 0.0 and part[y2][x2] == 0.0: part_lines.append(line) if is_single and isIndependentCropLines(part_lines): strokes.append(part) used_components.append(i) print("single stroke num: %d" % len(strokes)) print("used components num: %d" % len(used_components)) # cluster components based on the cropping lines for i in range(len(components)): if i in used_components: continue # find corresponding crop lines lines_id = [] for j in range(len(crop_lines)): line = crop_lines[j] if components[i][line[0][1]][line[0][0]] == 0.0 and components[i][line[1][1]][line[1][0]] != 0.0: lines_id.append(j) if components[i][line[0][1]][line[0][0]] != 0.0 and components[i][line[1][1]][line[1][0]] == 0.0: lines_id.append(j) if len(lines_id) == 0: continue component_line_relation[i] = lines_id used_components.append(i) # cluster components based on the relations and merge those related components clusters = [] for k1, v1 in component_line_relation.items(): cluster = [k1]; value_sets = [set(v1)] for k2, v2 in component_line_relation.items(): is_related = True for value in value_sets: if not value.intersection(set(v2)): is_related = False break if is_related and k2 not in cluster: cluster.append(k2) value_sets.append(set(v2)) cluster = sorted(cluster) if cluster not in clusters: clusters.append(cluster) print(clusters) # merge components based on the cluster for i in range(len(clusters)): cluster = clusters[i] bk = createBlankGrayscaleImage(image) for clt in cluster: bk = mergeBkAndComponent(bk, components[clt]) # add to strokes strokes.append(bk) # check the stroke is valid for i in range(len(strokes)): stroke = strokes[i] cv2.imshow("radical_%d" % index, contour_rgb) cv2.imshow("radical_gray_%d" % index, contour_gray) return strokes
def main(): # 0107亻 1133壬 0554十 0427凹 path = "0554十.jpg" # open image img = cv2.imread(path, 0) _, img = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY) img_rgb = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB) # contour without break points contour = getContourOfImage(img.copy()) contour = removeBreakPointsOfContour(contour) contour_rgb = cv2.cvtColor(contour, cv2.COLOR_GRAY2RGB) contours = splitConnectedComponents(contour) print("contours num: %d" % len(contours)) contours_sorted = [] for cont in contours: points = sortPointsOnContourOfImage(cont) print("points num: %d" % len(points)) contours_sorted.append(points) contour_points = [] for y in range(contour.shape[0]): for x in range(contour.shape[1]): if contour[y][x] == 0.0: # black points contour_points.append((x, y)) print("contour points num:%d" % len(contour_points)) # skeleton without extra branches skeleton = getSkeletonOfImage(img.copy()) # remove extra branches end_points = getEndPointsOfSkeletonLine(skeleton) cross_points = getCrossPointsOfSkeletonLine(skeleton) print("originale end: %d and cross: %d" % (len(end_points), len(cross_points))) skeleton_nobranches = removeBranchOfSkeletonLine(skeleton.copy(), end_points, cross_points) skeleton = skeleton_nobranches # new end points and cross points end_points = getEndPointsOfSkeletonLine(skeleton) cross_points = getCrossPointsOfSkeletonLine(skeleton) cross_points_bk = cross_points.copy() # merge the close points cross_points_merged = [] cross_distance_threshold = 10 used_index = [] for i in range(len(cross_points)): if i in used_index: continue pt1 = cross_points[i] midd_pt = None used_index.append(i) for j in range(len(cross_points)): if i == j or j in used_index: continue pt2 = cross_points[j] dist = math.sqrt((pt2[0] - pt1[0])**2 + (pt2[1] - pt1[1])**2) if dist < cross_distance_threshold: used_index.append(j) offset = (pt1[0] - pt2[0], pt1[1] - pt2[1]) print(offset) midd_pt = (pt2[0] + int(offset[0] / 2.), pt2[1] + int(offset[1] / 2.0)) if skeleton[midd_pt[1]][midd_pt[0]] == 0.0: cross_points_merged.append(midd_pt) else: min_distance = 100000000 current_pt = None for y in range(skeleton.shape[0]): for x in range(skeleton.shape[1]): if skeleton[y][x] == 0: dist = math.sqrt((midd_pt[0] - x)**2 + (midd_pt[1] - y)**2) if dist < min_distance: min_distance = dist current_pt = (x, y) if current_pt: cross_points_merged.append(current_pt) print("After merge cross points num: %d" % len(cross_points_merged)) cross_points = cross_points_merged print("After end: %d and cross: %d" % (len(end_points), len(cross_points))) skeleton_rgb = cv2.cvtColor(skeleton, cv2.COLOR_GRAY2RGB) # display all end points for pt in end_points: skeleton_rgb[pt[1]][pt[0]] = (0, 0, 255) for pt in cross_points: skeleton_rgb[pt[1]][pt[0]] = (0, 255, 0) for pt in cross_points_bk: skeleton_rgb[pt[1]][pt[0]] = (0, 0, 255) # all corner points on contour img = np.float32(img.copy()) dst = cv2.cornerHarris(img, 3, 3, 0.03) dst = cv2.dilate(dst, None) corners_area_points = [] for y in range(dst.shape[0]): for x in range(dst.shape[1]): if dst[y][x] > 0.1 * dst.max(): corners_area_points.append((x, y)) # show the corner points for pt in corners_area_points: if img[pt[1]][pt[0]] == 0: img_rgb[pt[1]][pt[0]] = (0, 255, 0) else: img_rgb[pt[1]][pt[0]] = (0, 0, 255) # all corner area points on the contour corners_lines_points = [] for pt in corners_area_points: if pt in contour_points: corners_lines_points.append(pt) for pt in corners_lines_points: contour_rgb[pt[1]][pt[0]] = (0, 255, 0) # merge points of corner points corners_merged_points = [] for contour_sorted in contours_sorted: i = 0 while True: midd_index = -1 pt = contour_sorted[i] if pt in corners_lines_points: # red point start = i end = start while True: end += 1 if end >= len(contour_sorted): break # next point next_pt = contour_sorted[end] if next_pt in corners_lines_points: # red point continue else: # black point break end -= 1 midd_index = start + int((end - start) / 2.0) i = end i += 1 if i >= len(contour_sorted): break if midd_index != -1: corners_merged_points.append(contour_sorted[midd_index]) print("After merged, corner points num: %d" % len(corners_merged_points)) for pt in corners_merged_points: contour_rgb[pt[1]][pt[0]] = (0, 0, 255) # remove the no-corner points corners_points = [] threshold_distance = 30 for pt in corners_merged_points: dist_cross = min_distance_point2pointlist(pt, cross_points) dist_end = min_distance_point2pointlist(pt, end_points) if dist_cross < threshold_distance and dist_end > threshold_distance / 3.: corners_points.append(pt) print("corner pints num: %d" % len(corners_points)) for pt in corners_points: contour_rgb[pt[1]][pt[0]] = (255, 0, 0) # segment contour to sub-contours based on the corner points def segmentContourBasedOnCornerPoints(contour_sorted, corner_points): """ Segment contour to sub-contours based on the corner points :param contour_sorted: :param corner_points: :return: """ if contour_sorted is None or corner_points is None: return # sub conotour index sub_contour_index = [] for pt in corner_points: index = contour_sorted.index(pt) sub_contour_index.append(index) print("sub contour index num: %d" % len(sub_contour_index)) sub_contours = [] for i in range(len(sub_contour_index)): if i == len(sub_contour_index) - 1: sub_contour = contour_sorted[sub_contour_index[i]:len( contour_sorted)] + contour_sorted[0:sub_contour_index[0] + 1] else: sub_contour = contour_sorted[ sub_contour_index[i]:sub_contour_index[i + 1] + 1] sub_contours.append(sub_contour) print("sub contours num: %d" % len(sub_contours)) return sub_contours # segment contour to sub-contours for contour in contours: cont_sorted = sortPointsOnContourOfImage(contour) sub_contours = segmentContourBasedOnCornerPoints( cont_sorted, corners_points) # cluster corner points corner_points_cluster = [] used_index = [] colinear_couple = [] for i in range(len(corners_points)): if i in used_index: continue for j in range(len(corners_points)): if i == j or j in used_index: continue min_offset = min(abs(corners_points[i][0] - corners_points[j][0]), abs(corners_points[i][1] - corners_points[j][1])) if min_offset < 20: couple = [corners_points[i], corners_points[j]] colinear_couple.append(couple) used_index.append(j) print("co linear num: %d" % len(colinear_couple)) print("sub contours num: %d" % len(sub_contours)) stroke1_img = np.ones_like(contour) * 255 stroke1_img = np.array(stroke1_img, dtype=np.uint8) stroke1_img_rgb = cv2.cvtColor(stroke1_img, cv2.COLOR_GRAY2RGB) for pt in sub_contours[0]: stroke1_img_rgb[pt[1]][pt[0]] = (0, 0, 0) stroke1_img[pt[1]][pt[0]] = 0 for pt in sub_contours[2]: stroke1_img_rgb[pt[1]][pt[0]] = (0, 0, 0) stroke1_img[pt[1]][pt[0]] = 0 cv2.line(stroke1_img_rgb, sub_contours[0][0], sub_contours[2][-1], (0, 0, 255), 1) cv2.line(stroke1_img_rgb, sub_contours[0][-1], sub_contours[2][0], (0, 0, 255), 1) cv2.line(stroke1_img, sub_contours[0][0], sub_contours[2][-1], 0, 1) cv2.line(stroke1_img, sub_contours[0][-1], sub_contours[2][0], 0, 1) stroke2_img = np.ones_like(contour) * 255 stroke2_img = np.array(stroke2_img, dtype=np.uint8) stroke2_img_rgb = cv2.cvtColor(stroke2_img, cv2.COLOR_GRAY2RGB) for pt in sub_contours[1]: stroke2_img_rgb[pt[1]][pt[0]] = (0, 0, 0) stroke2_img[pt[1]][pt[0]] = 0 for pt in sub_contours[3]: stroke2_img_rgb[pt[1]][pt[0]] = (0, 0, 0) stroke2_img[pt[1]][pt[0]] = 0 cv2.line(stroke2_img_rgb, sub_contours[1][0], sub_contours[3][-1], (0, 0, 255), 1) cv2.line(stroke2_img_rgb, sub_contours[1][-1], sub_contours[3][0], (0, 0, 255), 1) cv2.line(stroke2_img, sub_contours[1][0], sub_contours[3][-1], 0, 1) cv2.line(stroke2_img, sub_contours[1][-1], sub_contours[3][0], 0, 1) storke1_points = sortPointsOnContourOfImage(stroke1_img) stroke2_points = sortPointsOnContourOfImage(stroke2_img) stroke1_img = np.ones_like(stroke1_img) * 255 stroke1_img = np.array(stroke1_img, dtype=np.uint8) storke1_points = np.array([storke1_points], "int32") cv2.fillPoly(stroke1_img, storke1_points, 0) stroke2_img = np.ones_like(stroke2_img) * 255 stroke2_img = np.array(stroke2_img, dtype=np.uint8) storke2_points = np.array([stroke2_points], "int32") cv2.fillPoly(stroke2_img, storke2_points, 0) # find corresponding sub-contours based on the co-linear couple # for sub in sub_contours: # pt1 = sub[0] # pt2 = sub[-1] # # couples = [] # for coup in colinear_couple: # if pt1 in coup or pt2 in coup: # # if 4 points, 2 points should be in same sub-contour # if pt1 in coup and pt2 in coup: # continue # couples.append(coup) # print("sub couples num: %d" % len(couples)) # cv2.imshow("img rgb", img_rgb) # cv2.imshow("skeleton", skeleton) # cv2.imshow("skeleton no branches", skeleton_nobranches ) cv2.imshow("skeleton rgb", skeleton_rgb) cv2.imshow("contour rgb", contour_rgb) cv2.imshow("stroke 1", stroke1_img) cv2.imshow("stroke 2", stroke2_img) cv2.imshow("stroke1rgb", stroke1_img_rgb) cv2.imshow("stroke2rgb", stroke2_img_rgb) # for i in range(len(contours)): # cv2.imshow("contour %d" % i, contours[i]) cv2.waitKey(0) cv2.destroyAllWindows()
print("end points num: %d ,and cross points num: %d" % (len(end_points), len(cross_points))) # 9. Get contour image of component comp_contours_img = getContourImage(components[2]) comp_contours_img_rbg = cv2.cvtColor(comp_contours_img, cv2.COLOR_GRAY2RGB) # 10. Detect the number of contours and return all contours comp_contours = getConnectedComponents(comp_contours_img, connectivity=8) print("contours num: %d" % len(comp_contours)) # 11. Get points on contours corners_points = [] for cont in comp_contours: cont = removeBreakPointsOfContour(cont) cont_sorted = sortPointsOnContourOfImage(cont) cont_points = rdp(cont_sorted, 5) corners_points += cont_points CORNER_CROSS_DIST_THRESHOLD = 20 corners_points_merged = [] for pt in corners_points: for cpt in cross_points: dist = math.sqrt((pt[0] - cpt[0])**2 + (pt[1] - cpt[1])**2) if dist < CORNER_CROSS_DIST_THRESHOLD: corners_points_merged.append(pt) break # for pt in cross_points_: # cv2.circle(comp_skeleton_rgb, pt, 2, (0,255,0), 3) for pt in cross_points:
def main(): # load image img_path = "../templates/stroke_dan.png" img = cv2.imread(img_path, 0) _, img = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY) img_rgb = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB) # get contour of image contour = getContourOfImage(img) contour_rgb = cv2.cvtColor(contour, cv2.COLOR_GRAY2RGB) # fix breaking points on the contour break_points = [] for y in range(1, contour.shape[0] - 1): for x in range(1, contour.shape[1] - 1): if contour[y][x] == 0.0: num_ = getNumberOfValidPixels(contour, x, y) if num_ == 1: print((x, y)) break_points.append((x, y)) if len(break_points) != 0: contour = cv2.line(contour, break_points[0], break_points[1], color=0, thickness=1) cv2.imshow("c", contour) # order the contour points contour_points_ordered = sortPointsOnContourOfImage(contour) # contour_points_counter_clockwise = order_points(contour, isClockwise=False) print("number of points in ordered contour: %d" % len(contour_points_ordered)) # print("counter clock: %d" % len(contour_points_counter_clockwise)) contour_rgb_clock = contour_rgb.copy() contour_smooth_rgb_clock = contour_rgb.copy() # contour_rgb_counter_clock = contour_rgb.copy() # get key points on contour corners = cv2.goodFeaturesToTrack(contour, 6, 0.01, 10) corners = np.int0(corners) print("number of key points on contour: %d" % len(corners)) index = 0 corner_points_ = [] for i in corners: MAX_DIST = 10000 x, y = i.ravel() pt_ = None if (x, y - 1) in contour_points_ordered: pt_ = (x, y - 1) elif (x + 1, y - 1) in contour_points_ordered: pt_ = (x + 1, y - 1) elif (x + 1, y) in contour_points_ordered: pt_ = (x + 1, y) elif (x + 1, y + 1) in contour_points_ordered: pt_ = (x + 1, y + 1) elif (x, y + 1) in contour_points_ordered: pt_ = (x, y + 1) elif (x - 1, y + 1) in contour_points_ordered: pt_ = (x - 1, y + 1) elif (x - 1, y) in contour_points_ordered: pt_ = (x - 1, y) elif (x - 1, y - 1) in contour_points_ordered: pt_ = (x - 1, y - 1) else: # find the nearest point on the contour minx = 0 miny = 0 for cp in contour_points_ordered: dist = math.sqrt((x - cp[0])**2 + (y - cp[1])**2) if dist < MAX_DIST: MAX_DIST = dist minx = cp[0] miny = cp[1] pt_ = (minx, miny) corner_points_.append(pt_) cv2.circle(contour_rgb, (pt_[0], pt_[1]), 1, (0, 0, 255), -1) cv2.putText(contour_rgb, str(index), (pt_[0], pt_[1]), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 2, cv2.LINE_AA) index += 1 print("orignal corner points number: %d" % len(corner_points_)) # order the corner points in the clockwise direction corner_points = [] index = 0 for pt in contour_points_ordered: if pt in corner_points_: corner_points.append(pt) cv2.circle(contour_rgb_clock, (pt[0], pt[1]), 3, (255, 0, 0), -1) cv2.putText(contour_rgb_clock, str(index), (pt[0], pt[1]), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 2, cv2.LINE_AA) index += 1 print("corner points len: %d" % len(corner_points)) # contour segmentation based on the corner points contour_lines = [] for id in range(len(corner_points)): start_point = corner_points[id] end_point = start_point if id == len(corner_points) - 1: end_point = corner_points[0] else: end_point = corner_points[id + 1] # contour segmentation contour_segmentation = [] start_index = contour_points_ordered.index(start_point) end_index = contour_points_ordered.index(end_point) if start_index <= end_index: # normal index contour_segmentation = contour_points_ordered[ start_index:end_index + 1] else: # end is at contour_segmentation = contour_points_ordered[start_index: len(contour_points_ordered)] + \ contour_points_ordered[0: end_index+1] contour_lines.append(contour_segmentation) print("number of contour segmentation: %d" % len(contour_lines)) # use different color to show the contour segmentation for id in range(len(contour_lines)): if id % 3 == 0: # red lines for pt in contour_lines[id]: contour_rgb_clock[pt[1]][pt[0]] = (0, 0, 255) elif id % 3 == 1: # blue line for pt in contour_lines[id]: contour_rgb_clock[pt[1]][pt[0]] = (255, 0, 0) elif id % 3 == 2: # green line for pt in contour_lines[id]: contour_rgb_clock[pt[1]][pt[0]] = (0, 255, 0) # original and smooth contour smoothed_contour_points = [] for id in range(len(contour_lines)): print("line index: %d" % id) # original contour for pt in contour_lines[id]: contour_smooth_rgb_clock[pt[1]][pt[0]] = (0, 0, 255) # smooth contour li_points = np.array(contour_lines[id]) beziers = fitCurve(li_points, maxError=30) print("len bezier: %d" % len(beziers)) # # print(beziers) for bez in beziers: print(len(bez)) bezier_points = draw_cubic_bezier(bez[0], bez[1], bez[2], bez[3]) for id in range(len(bezier_points) - 1): start_pt = bezier_points[id] end_pt = bezier_points[id + 1] cv2.line(contour_smooth_rgb_clock, start_pt, end_pt, (255, 0, 0)) smoothed_contour_points += bezier_points # fill color in contour with sorted smooth contour points print(len(smoothed_contour_points)) smoothed_contour_points = np.array([smoothed_contour_points], "int32") fill_contour_smooth = np.ones(img.shape) * 255 fill_contour_smooth = np.array(fill_contour_smooth, dtype=np.uint8) fill_contour_smooth = cv2.fillPoly(fill_contour_smooth, smoothed_contour_points, 0) cv2.imshow("src", img) cv2.imshow("contour", contour) # cv2.imshow("corners", contour_rgb) cv2.imshow("contour clock", contour_rgb_clock) cv2.imshow("smooth contour clock", contour_smooth_rgb_clock) # cv2.imshow("contour counter clock", contour_rgb_counter_clock) cv2.imshow("fill contour", fill_contour_smooth) cv2.waitKey(0) cv2.destroyAllWindows()
from utils.Functions import getContourOfImage, sortPointsOnContourOfImage, removeBreakPointsOfContour, \ getSkeletonOfImage, removeBranchOfSkeletonLine, getEndPointsOfSkeletonLine, \ getCrossPointsOfSkeletonLine, getNumberOfValidPixels # 0107亻 1133壬 0554十 path = "0107亻.jpg" img = cv2.imread(path, 0) _, img = cv2.threshold(img, 127, 255, cv2.THRESH_BINARY) img_copy = img.copy() contour = getContourOfImage(img) contour = removeBreakPointsOfContour(contour) contour_sorted = sortPointsOnContourOfImage(contour) print(len(contour_sorted)) print(contour_sorted[0]) print(contour_sorted[-1]) img = np.float32(img) dst = cv2.cornerHarris(img, 3, 3, 0.05) dst = cv2.dilate(dst, None) img_rgb = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB) contour_rgb = cv2.cvtColor(contour, cv2.COLOR_GRAY2RGB) contour_rgb1 = contour_rgb.copy() print(dst.shape) corner_points = [] for y in range(dst.shape[0]):
def autoSmoothContoursOfComponent(component, eplison=10, max_error=200): """ Automatically smooth the contours of component. :param component: :return: """ if component is None: return # 5. Get contours of this component component_contours = getContourOfImage(component) contours = getConnectedComponents(component_contours, connectivity=8) print("contours num: ", len(contours)) # 6. Process contours to get closed and 1-pixel width contours by removing break points contours_processed = [] for cont in contours: cont = removeBreakPointsOfContour(cont) contours_processed.append(cont) print("contours processed num: %d" % len(contours_processed)) # 7. Smooth contours with RDP and cubic bezeir fit curve contours_smoothed = [] for cont in contours_processed: cont_smoothed = [] # sorted points on contour cont_sorted = sortPointsOnContourOfImage(cont) # simplify contour with RDP cont_simp = rdp(cont_sorted, eplison) print("cont simp num: ", len(cont_simp)) # split contour into sub-contours for i in range(len(cont_simp) - 1): start_pt = cont_simp[i] end_pt = cont_simp[i + 1] start_index = cont_sorted.index(start_pt) end_index = cont_sorted.index(end_pt) sub_cont_points = np.array(cont_sorted[start_index:end_index + 1]) beziers = fitCurve(sub_cont_points, maxError=max_error) for bez in beziers: bezier_points = draw_cubic_bezier(bez[0], bez[1], bez[2], bez[3]) cont_smoothed += bezier_points contours_smoothed.append(cont_smoothed) print("contours smoothed num: ", len(contours_smoothed)) # fill black color in contour area if len(contours_smoothed) == 1: # no hole exist, directly fill black in the contour cont = contours_smoothed[0] # cont_points = sortPointsOnContourOfImage(cont) cont_points = np.array([cont], "int32") fill_contour_smooth = np.ones_like(component) * 255 fill_contour_smooth = np.array(fill_contour_smooth, dtype=np.uint8) fill_contour_smooth = cv2.fillPoly(fill_contour_smooth, cont_points, 0) return fill_contour_smooth else: # exist hole, should processed print("there are holes!") fill_img_list = [] hole_points = [] for cont in contours_smoothed: # cont_points = sortPointsOnContourOfImage(cont) cont_points = np.array([cont], "int32") fill_contour_smooth = np.ones_like(component) * 255 fill_contour_smooth = np.array(fill_contour_smooth, dtype=np.uint8) fill_contour_smooth = cv2.fillPoly(fill_contour_smooth, cont_points, 0) valid_num = same_num = 0 for y in range(component.shape[0]): for x in range(component.shape[1]): if component[y][x] == 0.0: valid_num += 1 if fill_contour_smooth[y][x] == 0.0: same_num += 1 if 1.0 * same_num / valid_num > 0.8: fill_img_list.append(fill_contour_smooth) print("ratio: %f" % (1.0 * same_num / valid_num)) else: print("ratio: %f" % (1.0 * same_num / valid_num)) for y in range(fill_contour_smooth.shape[0]): for x in range(fill_contour_smooth.shape[1]): if fill_contour_smooth[y][x] == 0.0: hole_points.append((x, y)) # merge filled images blank_temp = np.ones_like(component) * 255 for fl in fill_img_list: for y in range(fl.shape[0]): for x in range(fl.shape[1]): if fl[y][x] == 0.0: blank_temp[y][x] = fl[y][x] # hole points for pt in hole_points: blank_temp[pt[1]][pt[0]] = 255 return blank_temp