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
|
/*
* Copyright (C) 2005 - 2013 René Rebe, ExactCODE GmbH
* (C) 2005 - 2007 Archivista GmbH, CH-8042 Zuerich
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; version 2. A copy of the GNU General
* Public License can be found in the file LICENSE.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANT-
* ABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
* Public License for more details.
*
* Alternatively, commercial licensing options are available from the
* copyright holder ExactCODE GmbH Germany.
*/
/* Short Description:
* Any color-space to b/w (1-bit) optimization. Applications involving
* huge amount of document archiving often prefer storing the data
* as (compressed) b/w data even in the age of TB storage as higher
* (color) image often explores the data volume exponentially.
*
* For this case this algorithm automatically determines a threshold
* and performs some optimizations in order to loose as few details
* as possible (e.g. retrieve hand notes, stamps, etc. but leave
* the background color, coffee sparkels, dust and folding lines
* out.
*
* While the first version only performed a unsharp mask and auto-
* thresholding, the latest version does perform a segmentation-pass,
* to optimize the performance (by leaving out empty areas) and
* to keep dithering of image areas.
*/
#include <cstdlib>
#include <cmath>
#include <iostream>
#include <iomanip>
#ifdef _MSC_VER
#include <vector>
#endif
#include "Image.hh"
#include "Colorspace.hh"
#include "Matrix.hh"
#include "optimize2bw.hh"
//#include "Timer.cc"
void optimize2bw (Image& image, int low, int high, int threshold,
int sloppy_threshold,
int radius, double standard_deviation)
{
// do nothing if already at s/w, ...
if (image.spp == 1 && image.bps == 1)
return;
/* Convert to RGB to gray.
If the threshold is to be determined automatically, use color info. */
const bool debug = false;
// color normalize on background color
// search for background color
{
std::vector<std::vector<unsigned int> > hist = histogram(image);
colorspace_by_name (image, "rgb8");
int lowest = 255, highest = 0, bg_r = 0, bg_g = 0, bg_b = 0;
for (int i = 0; i <= 255; i++)
{
int r, g, b;
r = g = b = hist[0][i];
if (hist.size() > 1) {
g = hist[1][i];
b = hist[2][i];
}
if (debug)
std::cout << i << ": "<< r << " " << g << " " << b << std::endl;
const int magic = 2; // magic denoise constant
if (r >= magic || g >= magic || b >= magic)
{
if (i < lowest)
lowest = i;
if (i > highest)
highest = i;
}
if (hist[0][i] > hist[0][bg_r])
bg_r = i;
if (hist.size() > 1) {
if (hist[1][i] > hist[1][bg_r])
bg_g = i;
if (hist[2][i] > hist[2][bg_r])
bg_b = i;
} else {
bg_g = bg_b = i;
}
}
highest = (int) (.21267 * bg_r + .71516 * bg_g + .07217 * bg_b);
if (false)
std::cerr << "lowest: " << lowest << ", highest: " << highest
<< ", back rgb: " << bg_r << " " << bg_g << " " << bg_b
<< std::endl;
const int min_delta = 128;
lowest = std::max (std::min (lowest, highest - min_delta), 0);
highest = std::min (std::max (highest, lowest + min_delta), 255);
if (low)
lowest = low;
if (high)
highest = high;
if (false)
std::cerr << "after limit and overwrite, lowest: " << lowest
<< ", highest: " << highest << std::endl;
signed int a = (255 * 256) / (highest - lowest);
signed int b = (-a * lowest);
if (false)
std::cerr << "a: " << (float) a / 256
<< " b: " << (float) b / 256 << std::endl;
uint8_t* it = image.getRawData();
uint8_t* end = image.getRawDataEnd();
uint8_t* it2 = it;
while (it != end) {
int _r = *it++;
int _g = *it++;
int _b = *it++;
_r = (_r * a + b) / 256;
_g = (_g * a + b) / 256;
_b = (_b * a + b) / 256;
// clip
_r = std::max (std::min (_r, 255), 0);
_g = std::max (std::min (_g, 255), 0);
_b = std::max (std::min (_b, 255), 0);
// on-the-fly convert to gray with associated weighting
*it2++ = (_r*28 + _g*59 + _b*11) / 100;
}
image.spp = 1; // converted data RGB8->GRAY8
image.setRawData();
}
// Convolution Matrix (unsharp mask a-like)
if (radius > 0)
{
// compute kernel (convolution matrix to move over the iamge)
// Utility::AutoTimer<Utility::Timer> timer ("convolution");
matrix_type divisor = 0;
float sd = standard_deviation;
#ifdef _MSC_VER
std::vector<matrix_type> matrix(radius+1);
std::vector<matrix_type> matrix_2(radius+1);
#else
matrix_type matrix[radius+1];
matrix_type matrix_2[radius+1];
#endif
for (int d = 0; d <= radius; ++d) {
matrix_type v = (matrix_type) (exp (-((float)d*d) / (2. * sd * sd)) );
matrix[d] = v;
divisor+=v;
if (d>0)
divisor+=v;
}
// normalize (will not work with integer matrix type !)
divisor=1.0/divisor;
for (int i=0; i<=radius; i++) {
matrix[i]*=divisor;
matrix_2[i]=-matrix[i];
}
decomposable_sym_convolution_matrix (image, &matrix[0], &matrix_2[0], radius, radius, 2.0);
}
}
|
