-
Notifications
You must be signed in to change notification settings - Fork 12
/
algorithms.cpp
225 lines (178 loc) · 8.45 KB
/
algorithms.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
#include "algorithms.hpp"
#include <vigra/convolution.hxx>
#include <vigra/linear_algebra.hxx>
using namespace vigra::linalg;
namespace sift {
namespace alg {
const vigra::MultiArray<2, f32_t> convolveWithGauss(const vigra::MultiArray<2, f32_t>& img,
f32_t sigma) {
vigra::Kernel1D<f32_t> filter;
filter.initGaussian(sigma);
vigra::MultiArray<2, f32_t> tmp(img.shape());
vigra::MultiArray<2, f32_t> result(img.shape());
separableConvolveX(img, tmp, filter);
separableConvolveY(tmp, result, filter);
return result;
}
const vigra::MultiArray<2, f32_t> reduceToNextLevel(const vigra::MultiArray<2, f32_t>& img,
f32_t sigma) {
// image size at current level
const vigra::Shape2 s((img.width()+ 1) / 2, (img.height() + 1) / 2);
// resize result image to appropriate size
vigra::MultiArray<2, f32_t> out(s);
// downsample smoothed image
resizeImageNoInterpolation(convolveWithGauss(img, sigma), out);
return out;
}
const vigra::MultiArray<2, f32_t> increaseToNextLevel(const vigra::MultiArray<2, f32_t>& img,
f32_t sigma) {
// image size at current level
const vigra::Shape2 s(img.width() * 2, img.height() * 2);
// resize result image to appropriate size
vigra::MultiArray<2, f32_t> out(s);
// downsample smoothed image
resizeImageNoInterpolation(convolveWithGauss(img, sigma), out);
return out;
}
const vigra::MultiArray<2, f32_t> dog(const vigra::MultiArray<2, f32_t>& lower,
const vigra::MultiArray<2, f32_t>& higher) {
vigra::MultiArray<2, f32_t> result(vigra::Shape2(lower.shape()));
for (u16_t x = 0; x < lower.shape(0); x++) {
for (u16_t y = 0; y < lower.shape(1); y++) {
const f32_t dif = higher(x, y) - lower(x, y);
// don't get negative values
result(x, y) = 128 + dif;
}
}
return result;
}
const vigra::Matrix<f32_t> foDerivative(const std::array<vigra::MultiArray<2, f32_t>, 3>& img,
const Point<u16_t, u16_t>& p) {
const f32_t dx = (img[1](p.x - 1, p.y) - img[1](p.x + 1, p.y)) / 2;
const f32_t dy = (img[1](p.x, p.y - 1) - img[1](p.x, p.y + 1)) / 2;
const f32_t ds = (img[0](p.x, p.y) - img[2](p.x, p.y)) / 2;
vigra::Matrix<f32_t> result(vigra::Shape2(3, 1));
result(0, 0) = dx;
result(1, 0) = dy;
result(2, 0) = ds;
return result;
}
const vigra::Matrix<f32_t> soDerivative(const std::array<vigra::MultiArray<2, f32_t>, 3>& img,
const Point<u16_t, u16_t>& p) {
const f32_t dxx = img[1](p.x + 1, p.y) + img[1](p.x - 1, p.y) - 2 * img[1](p.x, p.y);
const f32_t dyy = img[1](p.x, p.y + 1) + img[1](p.x, p.y - 1) - 2 * img[1](p.x, p.y);
const f32_t dss = img[2](p.x, p.y) + img[0](p.x, p.y) - 2 * img[1](p.x, p.y);
const f32_t dxy = (img[1](p.x + 1, p.y + 1) - img[1](p.x - 1, p.y + 1) - img[1](p.x + 1, p.y - 1)
+ img[1](p.x - 1, p.y - 1)) / 2;
const f32_t dxs = (img[2](p.x + 1, p.y) - img[2](p.x - 1, p.y)
- img[0](p.x + 1, p.y) + img[0](p.x - 1, p.y)) / 2;
f32_t dys = (img[2](p.x, p.y + 1) - img[2](p.x, p.y + 1)
- img[0](p.x, p.y + 1) + img[0](p.x, p.y - 1)) / 2;
vigra::MultiArray<2, f32_t> sec_deriv(vigra::Shape2(3, 3));
sec_deriv(0, 0) = dxx;
sec_deriv(1, 0) = dxy;
sec_deriv(2, 0) = dxs;
sec_deriv(0, 1) = dxy;
sec_deriv(1, 1) = dyy;
sec_deriv(2, 1) = dys;
sec_deriv(0, 2) = dxs;
sec_deriv(1, 2) = dys;
sec_deriv(2, 2) = dss;
return sec_deriv;
}
f32_t gradientMagnitude(const vigra::MultiArray<2, f32_t>& img, const Point<u16_t, u16_t>& p) {
return std::sqrt(std::pow(img(p.x + 1, p.y) - img(p.x - 1, p.y), 2) +
std::pow(img(p.x, p.y + 1) - img(p.x, p.y - 1), 2));
}
f32_t gradientOrientation(const vigra::MultiArray<2, f32_t>& img, const Point<u16_t, u16_t>& p) {
const f32_t result = std::atan2(img(p.x, p.y + 1) - img(p.x, p.y - 1), img(p.x + 1, p.y) - img(p.x - 1, p.y));
return std::fmod(result + 360, 360);
}
const std::array<f32_t, 36> orientationHistogram36(
const vigra::MultiArray<2, f32_t>& orientations,
const vigra::MultiArray<2, f32_t>& magnitudes,
const vigra::MultiArray<2, f32_t>& current_gauss) {
std::array<f32_t, 36> bins = {{0}};
for (u16_t x = 0; x < orientations.width(); x++) {
for (u16_t y = 0; y < orientations.height(); y++) {
const f32_t sum = magnitudes(x, y) * current_gauss(x, y);
u16_t i = std::floor(orientations(x, y) / 10);
i = i % 35;
bins[i] += sum;
}
}
return bins;
}
const std::vector<f32_t> orientationHistogram8(
const vigra::MultiArray<2, f32_t>& orientations,
const vigra::MultiArray<2, f32_t>& magnitudes,
const vigra::MultiArray<2, f32_t>& current_gauss) {
std::vector<f32_t> bins(8, 0);
for (u16_t x = 0; x < orientations.width(); x++) {
for (u16_t y = 0; y < orientations.height(); y++) {
const f32_t sum = magnitudes(x, y) * current_gauss(x, y);
u16_t i = std::floor(orientations(x, y) / 45);
i = i % 7;
bins[i] += sum;
}
}
return bins;
}
f32_t vertexParabola(const Point<u16_t, f32_t>& ln, const Point<u16_t, f32_t>& peak,
const Point<u16_t, f32_t>& rn) {
vigra::MultiArray<2, f32_t> a(vigra::Shape2(3, 3));
a(0, 0) = std::pow(ln.x, 2);
a(1, 0) = std::pow(peak.x, 2);
a(2, 0) = std::pow(rn.x, 2);
a(0, 1) = ln.x;
a(1, 1) = peak.x;
a(2, 1) = rn.x;
a(0, 2) = 0;
a(1, 2) = 0;
a(2, 2) = 0;
vigra::MultiArray<2, f32_t> b(vigra::Shape2(3, 1));
b(0, 0) = ln.y;
b(1, 0) = peak.y;
b(2, 0) = rn.y;
vigra::MultiArray<2, f32_t> res(vigra::Shape2(3, 1));
linearSolve(a, b, res);
return -res(1, 0) / (2 * res(0, 0));
}
std::array<Point<f32_t, f32_t>, 4> rotateShape(const Point<u16_t, u16_t>& center, f32_t angle,
const u16_t width, const u16_t height) {
//Determine upper left and bottom right point
auto ul = Point<f32_t, f32_t>(center.x - width / 2, center.y - height / 2);
auto ur = Point<f32_t, f32_t>(center.x - width / 2, center.y + height / 2);
auto bl = Point<f32_t, f32_t>(center.x + width / 2, center.y - height / 2);
auto br = Point<f32_t, f32_t>(center.x + width / 2, center.y + height / 2);
std::array<Point<f32_t, f32_t>, 4> shape{{ul, ur, bl, br}};
//Perform clockwise rotation
angle *= -1;
for (Point<f32_t, f32_t>& p : shape) {
//Transform center to 0, 0
p.x -= center.x;
p.y -= center.y;
//Perform rotation based on rotation matrix for both points
p.x = ul.x * std::cos(angle) - p.y * std::sin(angle);
p.y = ul.x * std::sin(angle) + p.y * std::cos(angle);
//Transform center back to original position
ul.x += center.x;
ul.y += center.y;
}
return shape;
}
void normalizeVector(std::vector<f32_t>& vec) {
//Get length of vector
f32_t length = 0;
std::for_each(vec.begin(), vec.end(), [&](f32_t& n) {
length += n;
});
if (length == 0) {
return;
}
std::for_each(vec.begin(), vec.end(), [&](f32_t& n) {
n /= length;
});
}
}
}