-
Notifications
You must be signed in to change notification settings - Fork 0
/
get_line_image_7.py
212 lines (163 loc) · 8.48 KB
/
get_line_image_7.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
import argparse
import os
import xml.dom.minidom as minidom
from PIL import Image
import numpy as np
from scipy.misc import toimage
from scipy.spatial import ConvexHull
from math import sqrt
from math import atan2, cos, sin, pi, degrees
from collections import namedtuple
bounding_box = namedtuple('bounding_box', ('area',
'length_parallel',
'length_orthogonal',
'rectangle_center',
'unit_vector',
'unit_vector_angle',
'corner_points'))
def unit_vector(pt0, pt1):
dis_0_to_1 = sqrt((pt0[0] - pt1[0])**2 + (pt0[1] - pt1[1])**2)
return (pt1[0] - pt0[0]) / dis_0_to_1, \
(pt1[1] - pt0[1]) / dis_0_to_1
def orthogonal_vector(vector):
return -1 * vector[1], vector[0]
def bounding_area(index, hull):
unit_vector_p = unit_vector(hull[index], hull[index+1])
unit_vector_o = orthogonal_vector(unit_vector_p)
dis_p = tuple(np.dot(unit_vector_p, pt) for pt in hull)
dis_o = tuple(np.dot(unit_vector_o, pt) for pt in hull)
min_p = min(dis_p)
min_o = min(dis_o)
len_p = max(dis_p) - min_p
len_o = max(dis_o) - min_o
return {'area': len_p * len_o,
'length_parallel': len_p,
'length_orthogonal': len_o,
'rectangle_center': (min_p + len_p / 2, min_o + len_o / 2),
'unit_vector': unit_vector_p,
}
def to_xy_coordinates(unit_vector_angle, point):
angle_orthogonal = unit_vector_angle + pi / 2
return point[0] * cos(unit_vector_angle) + point[1] * cos(angle_orthogonal), \
point[0] * sin(unit_vector_angle) + point[1] * sin(angle_orthogonal)
def rotate_points(center_of_rotation, angle, points):
rot_points = []
ang = []
for pt in points:
diff = tuple([pt[d] - center_of_rotation[d] for d in range(2)])
diff_angle = atan2(diff[1], diff[0]) + angle
ang.append(diff_angle)
diff_length = sqrt(sum([d**2 for d in diff]))
rot_points.append((center_of_rotation[0] + diff_length * cos(diff_angle),
center_of_rotation[1] + diff_length * sin(diff_angle)))
return rot_points
def rectangle_corners(rectangle):
corner_points = []
for i1 in (.5, -.5):
for i2 in (i1, -1 * i1):
corner_points.append((rectangle['rectangle_center'][0] + i1 * rectangle['length_parallel'],
rectangle['rectangle_center'][1] + i2 * rectangle['length_orthogonal']))
return rotate_points(rectangle['rectangle_center'], rectangle['unit_vector_angle'], corner_points)
def minimum_bounding_box(points):
if len(points) <= 2: raise ValueError('More than two points required.')
hull_ordered = [points[index] for index in ConvexHull(points).vertices]
hull_ordered.append(hull_ordered[0])
hull_ordered = tuple(hull_ordered)
min_rectangle = bounding_area(0, hull_ordered)
for i in range(1, len(hull_ordered)-1):
rectangle = bounding_area(i, hull_ordered)
if rectangle['area'] < min_rectangle['area']:
min_rectangle = rectangle
min_rectangle['unit_vector_angle'] = atan2(min_rectangle['unit_vector'][1], min_rectangle['unit_vector'][0])
min_rectangle['rectangle_center'] = to_xy_coordinates(min_rectangle['unit_vector_angle'], min_rectangle['rectangle_center'])
return bounding_box(
area=min_rectangle['area'],
length_parallel=min_rectangle['length_parallel'],
length_orthogonal=min_rectangle['length_orthogonal'],
rectangle_center=min_rectangle['rectangle_center'],
unit_vector=min_rectangle['unit_vector'],
unit_vector_angle=min_rectangle['unit_vector_angle'],
corner_points=set(rectangle_corners(min_rectangle))
)
def get_center(im):
center_x = im.size[0]/2
center_y = im.size[1]/2
return center_x, center_y
def get_horizontal_angle(unit_vector_angle):
if unit_vector_angle > pi / 2 and unit_vector_angle <= pi:
unit_vector_angle = unit_vector_angle - pi
elif unit_vector_angle > -pi and unit_vector_angle < -pi / 2:
unit_vector_angle = unit_vector_angle + pi
return unit_vector_angle
def get_smaller_angle(bounding_box):
unit_vector = bounding_box.unit_vector
unit_vector_angle = bounding_box.unit_vector_angle
ortho_vector = orthogonal_vector(unit_vector)
ortho_vector_angle = atan2(ortho_vector[1], ortho_vector[0])
unit_vector_angle_updated = get_horizontal_angle(unit_vector_angle)
ortho_vector_angle_updated = get_horizontal_angle(ortho_vector_angle)
if abs(unit_vector_angle_updated) < abs(ortho_vector_angle_updated):
return unit_vector_angle_updated
else:
return ortho_vector_angle_updated
def rotated_points(bounding_box, center):
p1, p2, p3, p4 = bounding_box.corner_points
x1, y1 = p1
x2, y2 = p2
x3, y3 = p3
x4, y4 = p4
center_x, center_y = center
rotation_angle_in_rad = -get_smaller_angle(bounding_box)
x_dash_1 = (x1 - center_x) * cos(rotation_angle_in_rad) - (y1 - center_y) * sin(rotation_angle_in_rad) + center_x
x_dash_2 = (x2 - center_x) * cos(rotation_angle_in_rad) - (y2 - center_y) * sin(rotation_angle_in_rad) + center_x
x_dash_3 = (x3 - center_x) * cos(rotation_angle_in_rad) - (y3 - center_y) * sin(rotation_angle_in_rad) + center_x
x_dash_4 = (x4 - center_x) * cos(rotation_angle_in_rad) - (y4 - center_y) * sin(rotation_angle_in_rad) + center_x
y_dash_1 = (y1 - center_y) * cos(rotation_angle_in_rad) + (x1 - center_x) * sin(rotation_angle_in_rad) + center_y
y_dash_2 = (y2 - center_y) * cos(rotation_angle_in_rad) + (x2 - center_x) * sin(rotation_angle_in_rad) + center_y
y_dash_3 = (y3 - center_y) * cos(rotation_angle_in_rad) + (x3 - center_x) * sin(rotation_angle_in_rad) + center_y
y_dash_4 = (y4 - center_y) * cos(rotation_angle_in_rad) + (x4 - center_x) * sin(rotation_angle_in_rad) + center_y
return x_dash_1, y_dash_1, x_dash_2, y_dash_2, x_dash_3, y_dash_3, x_dash_4, y_dash_4
def set_line_image_data(image, line_id, image_file_name):
base_name = os.path.splitext(os.path.basename(image_file_name))[0]
line_image_file_name = base_name + line_id + '.tif'
imgray = image.convert('L')
imgray_rev_arr = np.fliplr(imgray)
imgray_rev = toimage(imgray_rev_arr)
imgray_rev.save(os.path.join(data_path, 'lines', line_image_file_name))
def get_line_images_from_page_image(image_file_name, madcat_file_path):
im = Image.open(image_file_name)
doc = minidom.parse(madcat_file_path)
zone = doc.getElementsByTagName('zone')
for node in zone:
id = node.getAttribute('id')
if id == 'z14':
token_image = node.getElementsByTagName('token-image')
minimum_bounding_box_input = []
for token_node in token_image:
word_point = token_node.getElementsByTagName('point')
col_word, row_word = [], []
for word_node in word_point:
col_word.append(int(word_node.getAttribute('x')))
row_word.append(int(word_node.getAttribute('y')))
word_coordinate = (int(word_node.getAttribute('x')), int(word_node.getAttribute('y')))
minimum_bounding_box_input.append(word_coordinate)
bounding_box = minimum_bounding_box(minimum_bounding_box_input)
rotation_angle_in_rad = get_smaller_angle(bounding_box)
img2 = im.rotate(degrees(rotation_angle_in_rad), resample=Image.BICUBIC)
x_dash_1, y_dash_1, x_dash_2, y_dash_2, x_dash_3, y_dash_3, x_dash_4, y_dash_4 = rotated_points(
bounding_box, get_center(im))
min_x = min(x_dash_1, x_dash_2, x_dash_3, x_dash_4)
min_y = min(y_dash_1, y_dash_2, y_dash_3, y_dash_4)
max_x = max(x_dash_1, x_dash_2, x_dash_3, x_dash_4)
max_y = max(y_dash_1, y_dash_2, y_dash_3, y_dash_4)
box = (min_x, min_y, max_x, max_y)
region = img2.crop(box)
set_line_image_data(region, id, image_file_name)
### main ###
data_path = '/Users/ashisharora/madcat_ar'
for file in os.listdir(os.path.join(data_path, 'images')):
if file.endswith(".tif"):
image_path = os.path.join(data_path, 'images', file)
gedi_file_path = os.path.join(data_path, 'madcat', file)
gedi_file_path = gedi_file_path.replace(".tif", ".madcat.xml")
get_line_images_from_page_image(image_path, gedi_file_path)