def remove_lines(lines, filtered_lines, scores):
new_horiz_lines = cut_lines(lines[0], filtered_lines[0], scores[0])
new_vert_lines = cut_lines(lines[1], filtered_lines[1], scores[1])
scorer.add_score('new_horiz_lines', len(new_horiz_lines))
scorer.add_score('new_vert_lines', len(new_vert_lines))
return (new_horiz_lines, new_vert_lines)
def get_structure(boxes, lines):
# rows = cluster_boxes(boxes, 1)
# cols = cluster_boxes(boxes, 0)
row_clusters, col_clusters = rate_combinations(boxes, lines)
rows = translate_clusters(row_clusters)
cols = translate_clusters(col_clusters)
sorted_rows = sorted(rows, key = lambda row: (row[1], row[7]))
sorted_cols = sorted(cols, key = lambda col: (col[0], col[6]))
scorer.add_score('initial_rows', len(sorted_rows))
scorer.add_score('initial_cols', len(sorted_cols))
# combined_rows = combine_overlapping_neighbors(sorted_rows, 1, 0.5)
# combined_cols = combine_overlapping_neighbors(sorted_cols, 0, 0.5)
# scorer.add_score('combined_rows', len(combined_rows))
# scorer.add_score('combined_cols', len(combined_cols))
# return (combined_rows, combined_cols)
return (sorted_rows, sorted_cols)
def rate_combinations(boxes, lines):
overall_row_scores = {}
row_score_matrix = [[1.0 for x in range(len(boxes))] for y in range(len(boxes))]
overall_col_scores = {}
col_score_matrix = [[1.0 for x in range(len(boxes))] for y in range(len(boxes))]
horiz_lines = lines[0]
vert_lines = lines[1]
for comb in combinations(enumerate(boxes), 2):
row_scores = {}
col_scores = {}
i = comb[0][0]
j = comb[1][0]
box_1 = {
'left': comb[0][1][0],
'right': comb[0][1][0] + comb[0][1][2],
'top': comb[0][1][1],
'bottom': comb[0][1][1] + comb[0][1][3]
}
box_2 = {
'left': comb[1][1][0],
'right': comb[1][1][0] + comb[1][1][2],
'top': comb[1][1][1],
'bottom': comb[1][1][1] + comb[1][1][3]
}
# 1.) Their vertical (horizontal) centers align
# May want to cut the factor down to 1.0 to make it a max of 1.0
row_scores['center_align'] = 2.0 / (1.0 + abs(box_1['top'] + box_1['bottom'] - box_2['top'] - box_2['bottom']))
col_scores['center_align'] = 2.0 / (1.0 + abs(box_1['left'] + box_1['right'] - box_2['left'] - box_2['right']))
# 2.) Their left (top) edges align
row_scores['top_align'] = 1.0 / (1.0 + abs(box_1['top'] - box_2['top']))
col_scores['left_align'] = 1.0 / (1.0 + abs(box_1['left'] - box_2['left']))
# 3.) Their right (bottom) edges align
row_scores['bottom_align'] = 1.0 / (1.0 + abs(box_1['bottom'] - box_2['bottom']))
col_scores['right_align'] = 1.0 / (1.0 + abs(box_1['right'] - box_2['right']))
# 4.) If there is a line close to their left (above them)
row_scores['top_line'] = calculate_preceding_line_score(box_1['top'], box_2['top'], horiz_lines)
col_scores['left_line'] = calculate_preceding_line_score(box_1['left'], box_2['left'], vert_lines)
# 5.) If there is a line close to their right (below them)
row_scores['bottom_line'] = calculate_succeeding_line_score(box_1['bottom'], box_2['bottom'], horiz_lines)
col_scores['right_line'] = calculate_succeeding_line_score(box_1['right'], box_2['right'], vert_lines)
# 6.) They overlap significantly in their horizontal (vertical) range
row_scores['vert_overlap'] = calculate_vertical_overlap(box_1, box_2)
col_scores['horiz_overlap'] = calculate_horizontal_overlap(box_1, box_2)
# 7.) I would like to add in a term regarding a shared strong score with a third object
row_score = calculate_row_score(row_scores)
col_score = calculate_col_score(col_scores)
overall_row_scores[str(comb)] = row_score
overall_col_scores[str(comb)] = col_score
row_score_matrix[comb[0][0]][comb[1][0]] = row_score
row_score_matrix[comb[1][0]][comb[0][0]] = row_score
col_score_matrix[comb[0][0]][comb[1][0]] = col_score
col_score_matrix[comb[1][0]][comb[0][0]] = col_score
# for comb in overall_row_scores:
# print('comb: ' + str(comb))
# print('row score: ' + str(overall_row_scores[comb]))
# print('col score: ' + str(overall_col_scores[comb]))
# Might want to do 0.999 later
row_clusters = clusterer.newer_cluster_scores(row_score_matrix, 1.0)
col_clusters = clusterer.newer_cluster_scores(col_score_matrix, 1.0)
# print('Row clusters found:')
# for cluster in row_clusters:
# print('*****************')
# for i in cluster:
# print(boxes[i])
# print('------')
# print('Col clusters found:')
# for cluster in col_clusters:
# print('*****************')
# for i in cluster:
# print(boxes[i])
scorer.add_score('cluster_rows', len(row_clusters))
scorer.add_score('cluster_cols', len(col_clusters))
# print('done clustering')
# Now translate the clusters of indexes into clusters of boxes
row_cluster_boxes = []
for row in row_clusters:
row_cluster_boxes.append([])
for box_index in row:
row_cluster_boxes[len(row_cluster_boxes) - 1].append(boxes[box_index])
col_cluster_boxes = []
for col in col_clusters:
col_cluster_boxes.append([])
for box_index in col:
col_cluster_boxes[len(col_cluster_boxes) - 1].append(boxes[box_index])
return (row_cluster_boxes, col_cluster_boxes)
def get_lines(img_name, base_path):
img = cv2.imread(base_path + '/' + img_name)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
edges = cv2.Canny(gray, 50, 150, apertureSize = 3)
# cv2.imwrite('regents/canny/' + img_name, edges)
# 120, 20, 10 is good. Also 80, 20, 1
lines = cv2.HoughLinesP(edges, 1, np.pi / 180, 120, minLineLength=40, maxLineGap=2)
if lines is None:
lines = []
horiz_count = 0
vert_count = 0
horiz_lines = []
vert_lines = []
for info in lines:
x1, y1, x2, y2 = info[0]
line_info = {}
if abs(y1 - y2) < 0.1:
# This is a horizontal line
line_info['border'] = int((y1 + y2) / 2)
line_info['start'] = x1
line_info['end'] = x2
horiz_lines.append(line_info)
horiz_count += 1
elif abs(x1 - x2) < 0.1:
# This is a vertical line
line_info['border'] = int((x1 + x2) / 2)
line_info['start'] = y1
line_info['end'] = y2
vert_lines.append(line_info)
vert_count += 1
elif verbose:
print('Nonstandard line: ' + str(theta))
scorer.add_score('horiz_lines', len(horiz_lines))
scorer.add_score('vert_lines', len(vert_lines))
scorer.add_score('line_outside_rows', len(horiz_lines) - 1)
scorer.add_score('line_outside_cols', len(vert_lines) - 1)
scorer.add_score('line_inside_rows', len(horiz_lines) + 1)
scorer.add_score('line_inside_cols', len(vert_lines) + 1)
return (horiz_lines, vert_lines)