-
Notifications
You must be signed in to change notification settings - Fork 0
/
kernel.cpp
229 lines (185 loc) · 5.27 KB
/
kernel.cpp
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
#include "kernel.h"
#include <cmath>
#include <cstring>
using namespace byteimage;
Kernel::Kernel() : nr(0), nc(0), values(nullptr) { }
Kernel::Kernel(int nr, int nc) : nr(nr), nc(nc) {
values = new double [nr * nc];
memset(values, 0, nr * nc * sizeof(double));
}
Kernel::Kernel(const Kernel& k) : nr(0), nc(0), values(nullptr) {*this = k;}
Kernel::Kernel(Kernel&& k) : nr(0), nc(0), values(nullptr) {*this = k;}
Kernel::~Kernel() {delete [] values;}
Kernel& Kernel::operator=(const Kernel& k) {
if (nr * nc != k.nr * k.nc) {
delete [] values;
values = new double [k.nr * k.nc];
}
nr = k.nr;
nc = k.nc;
memcpy(values, k.values, nr * nc * sizeof(double));
return *this;
}
Kernel& Kernel::operator=(Kernel&& k) {
nr = k.nr; k.nr = 0;
nc = k.nc; k.nc = 0;
values = k.values; k.values = nullptr;
return *this;
}
double Kernel::boundedAt(int i, int j) const {
if (i + nr / 2 < 0 || i + nr / 2 >= nr
|| j + nc / 2 < 0 || j + nc / 2 >= nc) {
return 0.0;
}
else return at(i, j);
}
double Kernel::min() const {
double sum = 0.0;
for (int i = 0; i < nr * nc; i++)
if (values[i] < 0.0) sum += values[i];
return sum;
}
double Kernel::max() const {
double sum = 0.0;
for (int i = 0; i < nr * nc; i++)
if (values[i] > 0.0) sum += values[i];
return sum;
}
Kernel Kernel::transpose() const {
Kernel k(nc, nr);
for (int r = 0; r < nr; r++)
for (int c = 0; c < nc; c++)
k.values[c * nr + r] = values[r * nc + c];
return k;
}
double Kernel::convolveAt(const ByteImage& I, int r, int c, int ch) const {
int r1, c1;
double sum = 0.0;
for (int i = -nr / 2; i <= nr / 2; i++) {
r1 = r + i;
#ifdef CONVOLVE_NO_EDGES
if (r1 < 0 || r1 >= I.nr) continue;
#else
if (r1 < 0) r1 = 0;
else if (r1 >= I.nr) r1 = I.nr - 1;
#endif
for (int j = -nc / 2; j <= nc / 2; j++) {
c1 = c + j;
#ifdef CONVOLVE_NO_EDGES
if (c1 < 0 || c1 >= I.nc) continue;
#else
if (c1 < 0) c1 = 0;
else if (c1 >= I.nc) c1 = I.nc - 1;
#endif
sum += I.at(r1, c1, ch) * at(i, j);
}
}
return sum;
}
ByteImage Kernel::convolve(const ByteImage& I) const {
ByteImage O(I.nr, I.nc, I.nchannels);
double min = this->min();
double scale = this->max() - min;
for (int ch = 0; ch < O.nchannels; ch++)
for (int r = 0; r < O.nr; r++)
for (int c = 0; c < O.nc; c++)
O.at(r, c, ch) = clip((convolveAt(I, r, c, ch) - min) / scale);
return O;
}
ByteImage Kernel::convolve(const ByteImage& I, double factor) const{
ByteImage O(I.nr, I.nc, I.nchannels);
for (int ch = 0; ch < O.nchannels; ch++)
for (int r = 0; r < O.nr; r++)
for (int c = 0; c < O.nc; c++)
O.at(r, c, ch) = clip(convolveAt(I, r, c, ch) * factor);
return O;
}
ByteImage Kernel::convolveSeparable(const ByteImage& I) const {
return convolve(transpose().convolve(I));
}
ByteImage Kernel::convolveMagnitude(const ByteImage& I) const {
ByteImage O(I.nr, I.nc, I.nchannels);
Kernel T = transpose();
const double C = 1.0 / sqrt(2);
double min = this->min();
double scale = this->max() - min;
double X, Y;
for (int ch = 0; ch < O.nchannels; ch++)
for (int r = 0; r < O.nr; r++)
for (int c = 0; c < O.nc; c++) {
X = (convolveAt(I, r, c, ch) - min) / scale;
Y = (T.convolveAt(I, r, c, ch) - min) / scale;
O.at(r, c, ch) = clip(C * sqrt(X * X + Y * Y));
}
return O;
}
void Kernel::print() {
for (int r = 0; r < nr; r++) {
printf("[\t");
for (int c = 0; c < nc; c++)
printf("%f\t", values[r * nc + c]);
printf("]\n");
}
}
Kernel Kernel::Identity() {
Kernel k(1, 1);
k.values[0] = 1.0;
return k;
}
Kernel Kernel::BoxBlur(int w) {
Kernel k(w, w);
for (int i = 0; i < w * w; i++)
k.values[i] = 1.0 / (w * w);
return k;
}
Kernel Kernel::SobelX() {
Kernel k(3, 3);
k.values[0] = k.values[6] = -1.0;
k.values[3] = -2.0;
k.values[2] = k.values[8] = 1.0;
k.values[5] = 2.0;
return k;
}
Kernel Kernel::SobelY() {
return SobelX().transpose();
}
Kernel Kernel::Gaussian(double sigma) {
const double pi = 3.14159265358979;
int radius = (int)ceil(2 * sigma * sqrt(2.0 * log(10)));
int diameter = 2 * radius + 1;
double s2 = sigma * sigma;
double s2inv = 1.0 / s2;
double fact = pow(2 * pi * s2, -0.5);
Kernel k(1, diameter);
for (int x = 0; x <= radius; x++)
k.at(0, x) = k.at(0, -x) = fact * exp(-0.5 * x * x * s2inv);
return k;
}
Kernel Kernel::Gradient(double sigma) {
Kernel k = Gaussian(sigma);
int radius = k.nc / 2;
double s2inv = 1.0 / (sigma * sigma);
for (int x = -radius; x <= radius; x++)
k.at(0, x) = -x * s2inv * k.at(0, x);
return k;
}
Kernel Kernel::LoG(double sigma) {
Kernel k = Gaussian(sigma);
int radius = k.nc / 2;
double s2inv = 1.0 / (sigma * sigma);
for (int x = 0; x <= radius; x++)
k.at(0, -x) = k.at(0, x) = s2inv * (x * x * s2inv - 1) * k.at(0, x);
return k;
}
Kernel Kernel::LoG2D(double sigma) {
const double pi = 3.14159265358979;
int radius = ceil(3 * sigma);
int diameter = 2 * radius + 1;
double s2 = sigma * sigma;
double fact = -1.0 / (pi * s2 * s2);
Kernel k(diameter, diameter);
for (int y = 0; y < radius; y++)
for (int x = 0; x < radius; x++)
k.at(y, x) = k.at(y, -x) = k.at(-y, x) = k.at(-y, -x) = fact * (1.0 - (x * x + y * y) / (2 * s2)) * exp(-(x * x + y * y) / (2 * s2));
return k;
}