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
213
214
215
|
#!/usr/bin/python
import math
import graphic
from misc import *
class hocr_parser:
'''Parse HTML-OCR files.
Right now, the only thing this class does is get_bounding_boxes.
'''
hocr = []
def __init__(self, fname):
self.hocr = file2array(fname)
def get_bounding_boxes(self):
'''Extracts the bounding boxes from the hocr description of each line of text
Returns an array of tuples.
'''
bboxes = []
p = re.compile(r"ocr_line.*bbox (\d+) (\d+) (\d+) (\d+)")
for l in self.hocr:
m = p.search(l)
if m:
bboxes.append(map(int, m.groups()))
return bboxes
class splitter():
'''splitter(image, **opts)
This class operates on images of text. It separates that image
into smaller images, each containing a picture of one line from
that text. It assumes somewhat parallel lines of dark text on a
light background with no images.
image is an image file spec, a pils image or a graphic object.
Pass in a **dictionary with options:
brightness = percent brightness to adjust before processing. Final image is unenhanced.
contrast = percent contrast to adjust to before processing. Final image is unenhanced.
tilt = the max degrees the text tilts from the horizon (optional)
'''
tiff = ''
txt = ''
opt={}
def __init__(self, image, **opt):
self.opt = set_defaults( {'brightness': 100,
'contrast' : 100,
'tilt': 4,
'level' : False},
opt)
self.im = graphic.graphic(image)
self.unenhanced = graphic.graphic(image)
def get_blank_line(self, j, h_high):
i = 0
while i <= h_high - j:
y = j + int(i + 0.5)
r = self.im.whitespace((0,j), (self.im.w,y))
if r > 80:
return (j, y)
i = (abs(i) + 0.5) * (-1 * ((i > 0) * 2 - 1))
return -1, -1
def get_cut_lines(self):
'''find whitepace by trying to draw horizontalish lines that do not hit
many black pixels'''
height_range = int(calc_right_tri_leg(self.opt['tilt'], self.im.w))
cut_line = []
last = 'white'
j = 0
while j <= self.im.h:
h_high = j+height_range
if h_high > self.im.h:
h_high = self.im.h
coords = self.get_blank_line(j, h_high)
if coords[0] != -1:
if last == 'black':
cut_line.append((0,coords[0], self.im.w, coords[1]))
j = j + 5
last = 'white'
else:
## cycle through blank lines faster
if last == 'white':
last = 'black'
j = j + 4
j = j + 3
return cut_line
def level_line(self, region):
if self.opt['level'] == False:
return region
## Whiteout everything except the line of text
b1, b2 = region.blackest_line()
h1, h2 = region.whitest_line(0,int(mean(b1,b2)))
h3, h4 = region.whitest_line(int(mean(b1,b2)), region.size[1])
## Rotate line of text parallel to horizon
s = slope((0,b1), (self.im.w,b2))
degrees = math.degrees(math.atan(s))
region = region.rotate(degrees, expand=1) #TODO: get a better rotate algorithm
region = region.trim(255) #TODO: use imagemagic's fuzzy trim?
return region
def do_cuts(self, cuts, **kwargs):
last = (0,0,self.im.w,0)
counter=0
print "do_cuts"
kwargs = set_defaults( {'draw' : False}, kwargs)
if kwargs['draw']:
draw = ImageDraw.Draw(self.im.im)
strips = []
while counter < len(cuts):
j = cuts[counter]
print counter
region = self.unenhanced.copy()
if kwargs['draw']:
draw = ImageDraw.Draw(region.im)
draw.polygon(((0,0),(0,last[1]),(last[2],last[3]),(last[2],0)), fill=255)
draw.polygon(((0,j[1]),(j[2],j[3]),(j[2],self.im.h), (0,self.im.h)), fill=255)
region = region.crop((min(last[0], last[2]), min(last[1],last[3]),
max(j[0], j[2]), max(j[1],j[3])))
region = self.level_line(region)
strips.append(region)
counter = counter + 1
last = j
return strips
def show_cuts(self, cuts):
t = self.im.copy().im
draw = ImageDraw.Draw(t)
for j in cuts:
draw.line(j)
t.save('cuts.png')
self.im.show()
sys.exit()
def get_bounding_boxes(self):
if 'hocr' in self.opt:
parser = hocr_parser(self.opt['hocr'])
return parser.get_bounding_boxes()
else:
cuts = self.get_cut_lines()
print cuts
return cuts
def cut_bbox(self, bbox):
'''bbox is a list of tuples. Each tuple has 4 elements. Each
element is a coordinate and the four coordinates are x,y of
upper left and x,y of lower right of the bounding box. '''
cuts=[]
last = (0,0,self.im.w,0)
counter=0
while counter < len(bbox):
box = bbox[counter]
region = self.im.copy()
draw = ImageDraw.Draw(region.im)
region = region.crop((box[0], box[1], box[2], box[3]))
region = self.level_line(region)
cuts.append(region)
counter = counter + 1
last = box
return cuts
def tune_bbox(self, bbox):
'''Tune a bbox so it does not cut off letters.'''
return bbox
def tune_bboxes(self, bbox):
'''Tune bboxes so they do not cut off letters.
bbox is a list of 4-tuples containing left, top, right, bottom.'''
ret = []
for b in bbox:
ret.append(self.tune_bbox(b))
return ret
def split(self):
self.im.enhance(self.opt['brightness'], self.opt['contrast'])
if 'hocr' in self.opt and self.opt['hocr']:
parser = hocr_parser(self.opt['hocr'])
bbox = parser.get_bounding_boxes()
bbox = self.tune_bboxes(bbox)
cuts = self.cut_bbox(bbox)
else:
cuts = self.do_cuts(self.get_cut_lines())
return cuts
|