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
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
| function m = xycut(img_file, varargin)
% XYCUT Decompose the image passed into a set of subrectangle segments
% using the top-down X-Y cut page segmentation algorithm.
%
% M = XYCUT(IMG_FILE, {H_THRESH, V_THRESH}) splits the page specified by
% IMG_FILE into segments recursively by making cuts into the most prominent
% valley in the horizontal and vertical directions at each step. This
% process bottoms out when the valleys are less than H_THRESH and V_THRESH
% pixels in length.
%
% The nx4 matrix M returned lists the left,top,bottom,right co-ordinates of
% each of the n segments created.
%
% H_THRESH and V_THRESH are optional, and if left unspecified H_THRESH
% defaults to: 40 and V_THRESH defaults to: 20
%
% If there is a problem at any point, an error is returned to the caller.
% CVS INFO %
%%%%%%%%%%%%
% $Id: xycut.m,v 1.6 2004-10-06 20:12:16 klaven Exp $
%
% REVISION HISTORY:
% $Log: xycut.m,v $
% Revision 1.6 2004-10-06 20:12:16 klaven
% Fixed a minor bug in xycut that occasionally added a region to the list incorrectly.
%
% Revision 1.5 2004/09/15 14:58:30 klaven
% *** empty log message ***
%
% Revision 1.4 2004/08/16 22:34:34 klaven
% New best_cuts algorithm core complete, and an implementation for xycuts. The best_cuts implementation of xycuts produces identical results to xycuts.m.
%
% Revision 1.3 2004/07/27 22:04:34 klaven
% First run at code to find the optimal parameters for the xycut algorithm
%
% Revision 1.2 2004/06/28 16:22:39 klaven
% *** empty log message ***
%
% Revision 1.1 2004/06/19 00:27:28 klaven
% Re-organizing files. Third step: re-add the files.
%
% Revision 1.8 2004/05/13 21:17:01 klaven
% Working with some new functions for evaluatng the segmentation.
%
% Revision 1.7 2004/04/30 01:07:04 klaven
% This is the version to use for my demo.
%
% Revision 1.6 2004/04/30 00:25:42 klaven
% xycut now uses a snap function that jumps back to the beginning of the current mark once the threshold is crossed. The threshold is now cumulative since the start of the most recent mark.
%
% Revision 1.5 2004/04/28 18:53:54 klaven
% Tweaking the elimination of whitespace from the xycut algorithm. In progress.
%
% Revision 1.4 2004/04/26 22:53:22 klaven
% Changed xycut to include wst. The wst is the White Space Threshold - the minimum fraction of black pixels that still counts as whitespace. Setting this to 0 leaves the algorithm the same as before.
%
% Revision 1.3 2004/04/22 16:51:04 klaven
% Assorted changes made while testing lr and knn on larger samples
%
% Revision 1.2 2003/08/12 22:21:42 scottl
% Changed default thresholds to 30 and 30.
%
% Revision 1.1 2003/08/01 22:01:36 scottl
% Initial revision.
%
% LOCAL VARS %
%%%%%%%%%%%%%%
% default horizontal threshold (if not passed above)
%ht = 40; % prefered ht for single-column layout
%ht = 40; % optimized ht for jmlr
% ht = 20; %prefered ht for double-column layout
ht = 45; %optimized ht for nips
%vt = 18; % default vertical threshold (if not passed above)
%vt = 16; % optimized vt for jmlr
vt = 16; % optimized vt for nips
%wst = 0.009; % minimum percent ink in whitespace to count as valley
wst = 0.0000001;
% first do some argument sanity checking on the argument passed
error(nargchk(1,3,nargin));
if (ischar(img_file));
p = imread(img_file);
else
p = img_file;
end;
%if iscell(img_file) | ~ ischar(img_file) | size(img_file,1) ~= 1
% error('IMG_FILE must contain a single string.');
%end
if nargin >= 2
ht = varargin{1};
if nargin == 3
vt = varargin{2};
end
end
% attempt open the file and read in its pixel data
%p = imread(img_file);
% determine the initial page bounding box co-ords
%x1 = 1;
%y1 = 1;
%x2 = size(p,2);
%y2 = size(p,1);
start_seg = [1 1 size(p,2) size(p,1)];
%start_seg = seg_snap(p,start_seg,wst);
x1 = start_seg(1);
y1 = start_seg(2);
x2 = start_seg(3);
y2 = start_seg(4);
% recursively segment the bounding box to create the list of segments
m = segment(p, x1, y1, x2, y2, ht, vt, wst);
% SUBFUNCITON DECLARATIONS %
%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function res = segment(p, x1, y1, x2, y2, ht, vt, wst)
% SEGMENT Recursive subfunction that segments the rectangle passed into
% smaller pieces using the XY cut algorithm.
oldseg = [x1 y1 x2 y2];
% start by determining the sum of all non-background pixels in the horizontal
% and vertical directions within the co-ord box passed -- note we must use 1-
% pixel value since background pixels are 1
hmeans = mean(1 - p(y1:y2, x1:x2));
vmeans = mean(1 - p(y1:y2, x1:x2), 2);
% determine the longest background valley (sum value = 0) run in both
% directions
[hrunlength, hrunpos] = long_valley(hmeans, wst);
[vrunlength, vrunpos] = long_valley(vmeans, wst);
if vrunlength > vt & vrunlength >= hrunlength
vrunstart = vrunpos - floor(vrunlength / 2);
vrunend = vrunpos + floor(vrunlength / 2);
% make a horizontal cut along the vertical midpoint
newseg1 = [x1 y1 x2 (y1 + vrunpos)];
newseg1 = seg_snap(p,newseg1,wst);
newseg2 = [x1 (y1 + vrunpos) x2 y2];
newseg2 = seg_snap(p,newseg2,wst);
if (rect_comes_before(newseg2, newseg1));
res = ...
[segment(p,newseg2(1),newseg2(2),newseg2(3),newseg2(4),ht,vt,wst); ...
segment(p,newseg1(1),newseg1(2),newseg1(3),newseg1(4),ht,vt,wst)];
else
res = ...
[segment(p,newseg1(1),newseg1(2),newseg1(3),newseg1(4),ht,vt,wst); ...
segment(p,newseg2(1),newseg2(2),newseg2(3),newseg2(4),ht,vt,wst)];
end;
%res = [segment(p, x1, y1, x2, (y1 + vrunpos), ht, vt, wst); ...
% segment(p, x1, (y1 + vrunpos), x2, y2, ht, vt, wst)];
elseif hrunlength > ht & hrunlength >= vrunlength
hrunstart = hrunpos - floor(hrunlength / 2);
hrunend = hrunpos + floor(hrunlength / 2);
% make a vertical cut along the horizontal midpoint
newseg1 = [x1 y1 (x1 + hrunpos) y2];
newseg1 = seg_snap(p,newseg1,wst);
newseg2 = [(x1 + hrunpos) y1 x2 y2];
newseg2 = seg_snap(p,newseg2,wst);
if (rect_comes_before(newseg2, newseg1));
res = ...
[segment(p,newseg2(1),newseg2(2),newseg2(3),newseg2(4),ht,vt,wst); ...
segment(p,newseg1(1),newseg1(2),newseg1(3),newseg1(4),ht,vt,wst)];
else
res = ...
[segment(p,newseg1(1),newseg1(2),newseg1(3),newseg1(4),ht,vt,wst); ...
segment(p,newseg2(1),newseg2(2),newseg2(3),newseg2(4),ht,vt,wst)];
end;
%res = [segment(p, x1, y1, (x1 + hrunpos), y2, ht, vt, wst); ...
% segment(p, (x1 + hrunpos), y1, x2, y2, ht, vt, wst)];
else
% non-recursive case, don't split anything
res = seg_snap(p,oldseg,wst);
%[x1,y1,x2,y2] = snap(p,x1, y1, x2, y2,wst);
%res = [x1,y1,x2,y2];
end
%function [left,top,right,bottom] = snap(pixels, x1, y1, x2, y2, wst)
%% loop over the 4 sides, trimming our current bounding box until we have met
%% our ink thresholds
%
%bg = 1; % default value for background pixels
%
%left=x1; top = y1; right=x2; bottom=y2;
%
%l_done = false;
%t_done = false;
%r_done = false;
%b_done = false;
%
%l_markstart = left;
%t_markstart = top;
%r_markstart = right;
%b_markstart = bottom;
%
%
%while ~ (l_done & t_done & r_done & b_done)
%
% if left >= right | top >= bottom
% warning('Did not find sufficient ink. Returning orig. subrectangle');
% return;
% end
%
% if ~ l_done
% count = sum(sum(bg - pixels(top:bottom,left)));
% if (count == 0)
% l_markstart = left;
% left = left + 1;
% else
% count = sum(sum((bg - pixels(top:bottom,l_markstart:left))>0));
% if count > (wst * (bottom - top + 1))
% left = l_markstart;
% l_done = true;
% else
% left = left + 1;
% end
% end
% end
%
% if ~ t_done
% count = sum(sum(bg - pixels(top,left:right)));
% if (count == 0)
% t_markstart = top;
% top = top + 1;
% else
% count = sum(sum((bg - pixels(t_markstart:top,left:right))>0));
% if count > (wst * (right - left + 1))
% top = t_markstart;
% t_done = true;
% else
% top = top + 1;
% end
% end
% end
%
% if ~ r_done
% count = sum(sum(bg - pixels(top:bottom,right)));
% if (count == 0)
% r_markstart = right;
% right = right - 1;
% else
% count = sum(sum((bg - pixels(top:bottom,right:r_markstart))>0));
% if count > (wst * (bottom - top + 1))
% right = r_markstart;
% r_done = true;
% else
% right = right - 1;
% end
% end
% end
%
% if ~ b_done
%
% count = sum(sum((bg - pixels(bottom,left:right))>0));
%
% if (count == 0)
% b_markstart = bottom;
% bottom = bottom - 1;
% else
% count = sum(sum((bg - pixels(bottom:b_markstart,left:right))>0));
% if count > (wst * (right - left + 1))
% bottom = b_markstart;
% b_done = true;
% else
% bottom = bottom - 1;
% end
% end
% end
%
%end %end while loop
function [longlength, mid] = long_valley(means, wst)
% LONG_VALLEY Subfunction that determines the length and midpoint position of
% the longest background valley (meanss value <=wst) run in the
% vector passed. If the vector is all non-zero, a length of 0 is
% returned and the mid is undefiend (NaN).
% wst: White Space Threshold (fraction of ink pixels considered whitespace)
s = 1;
e = length(means);
currlength = 0;
longlength = 0;
mid = nan;
% strip the leading and trailing 0 runs from sums (since we want a full valley
% for our run count).
while (s < e) && (means(s) <= wst)
s = s + 1;
end
while (s < e) && (means(e) <= wst)
e = e - 1;
end
if s >= e
return;
end
for i = s:e
if means(i) <= wst
currlength = currlength + 1;
if currlength > longlength
longlength = currlength;
mid = i - floor(longlength / 2);
end
else
currlength = 0;
end
end |
