forked from sattarab/image-quality-tools
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathVSI.m
279 lines (238 loc) · 10.3 KB
/
VSI.m
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
function sim = VSI(image1, image2)
% ========================================================================
% Visual Saliency based Index
% Copyright(c) 2013 Lin ZHANG
% All Rights Reserved.
% ----------------------------------------------------------------------
% Permission to use, copy, or modify this software and its documentation
% for educational and research purposes only and without fee is hereQ
% granted, provided that this copyright notice and the original authors'
% names appear on all copies and supporting documentation. This program
% shall not be used, rewritten, or adapted as the basis of a commercial
% software or hardware product without first obtaining permission of the
% authors. The authors make no representations about the suitability of
% this software for any purpose. It is provided "as is" without express
% or implied warranty.
%----------------------------------------------------------------------
% For more information, please refer to
% Lin Zhang et al., "VSI: A Visual Saliency Induced Index for Perceptual
% Image Quality Assessment", submitted to TIP
%----------------------------------------------------------------------
%
%Input : (1) image1: the first image being compared, which is a RGB image
% (2) image2: the second image being compared, which is a RGB image
%
%Output: sim: the similarity score between two images, a real number
%
%-----------------------------------------------------------------------
constForVS = 1.27;%fixed
constForGM = 386;%fixed
constForChrom = 130;%fixed
alpha = 0.40;%fixed
lambda = 0.020;%fixed
sigmaF = 1.34;%fixed donot change
omega0 = 0.0210;%fixed
sigmaD = 145;%fixed
sigmaC = 0.001;%fixed
%compute the visual saliency map using SDSP
saliencyMap1 = SDSP(image1,sigmaF,omega0,sigmaD,sigmaC);
saliencyMap2 = SDSP(image2,sigmaF,omega0,sigmaD,sigmaC);
[rows, cols, junk] = size(image1);
%transform into an opponent color space
L1 = 0.06 * double(image1(:,:,1)) + 0.63 * double(image1(:,:,2)) + 0.27 * double(image1(:,:,3));
L2 = 0.06 * double(image2(:,:,1)) + 0.63 * double(image2(:,:,2)) + 0.27 * double(image2(:,:,3));
M1 = 0.30 * double(image1(:,:,1)) + 0.04 * double(image1(:,:,2)) - 0.35 * double(image1(:,:,3));
M2 = 0.30 * double(image2(:,:,1)) + 0.04 * double(image2(:,:,2)) - 0.35 * double(image2(:,:,3));
N1 = 0.34 * double(image1(:,:,1)) - 0.60 * double(image1(:,:,2)) + 0.17 * double(image1(:,:,3));
N2 = 0.34 * double(image2(:,:,1)) - 0.60 * double(image2(:,:,2)) + 0.17 * double(image2(:,:,3));
%%%%%%%%%%%%%%%%%%%%%%%%%
% Downsample the image
%%%%%%%%%%%%%%%%%%%%%%%%%
minDimension = min(rows,cols);
F = max(1,round(minDimension / 256));
aveKernel = fspecial('average',F);
aveM1 = conv2(M1, aveKernel,'same');
aveM2 = conv2(M2, aveKernel,'same');
M1 = aveM1(1:F:rows,1:F:cols);
M2 = aveM2(1:F:rows,1:F:cols);
aveN1 = conv2(N1, aveKernel,'same');
aveN2 = conv2(N2, aveKernel,'same');
N1 = aveN1(1:F:rows,1:F:cols);
N2 = aveN2(1:F:rows,1:F:cols);
aveL1 = conv2(L1, aveKernel,'same');
aveL2 = conv2(L2, aveKernel,'same');
L1 = aveL1(1:F:rows,1:F:cols);
L2 = aveL2(1:F:rows,1:F:cols);
aveSM1 = conv2(saliencyMap1, aveKernel,'same');
aveSM2 = conv2(saliencyMap2, aveKernel,'same');
saliencyMap1 = aveSM1(1:F:rows,1:F:cols);
saliencyMap2 = aveSM2(1:F:rows,1:F:cols);
%%%%%%%%%%%%%%%%%%%%%%%%%
% Calculate the gradient map
%%%%%%%%%%%%%%%%%%%%%%%%%
dx = [3 0 -3; 10 0 -10; 3 0 -3]/16;
dy = [3 10 3; 0 0 0; -3 -10 -3]/16;
IxL1 = conv2(L1, dx, 'same');
IyL1 = conv2(L1, dy, 'same');
gradientMap1 = sqrt(IxL1.^2 + IyL1.^2);
IxL2 = conv2(L2, dx, 'same');
IyL2 = conv2(L2, dy, 'same');
gradientMap2 = sqrt(IxL2.^2 + IyL2.^2);
%%%%%%%%%%%%%%%%%%%%%%%%%
% Calculate the VSI
%%%%%%%%%%%%%%%%%%%%%%%%%
VSSimMatrix = (2 * saliencyMap1 .* saliencyMap2 + constForVS) ./ (saliencyMap1.^2 + saliencyMap2.^2 + constForVS);
gradientSimMatrix = (2*gradientMap1.*gradientMap2 + constForGM) ./(gradientMap1.^2 + gradientMap2.^2 + constForGM);
weight = max(saliencyMap1, saliencyMap2);
ISimMatrix = (2 * M1 .* M2 + constForChrom) ./ (M1.^2 + M2.^2 + constForChrom);
QSimMatrix = (2 * N1 .* N2 + constForChrom) ./ (N1.^2 + N2.^2 + constForChrom);
SimMatrixC = (gradientSimMatrix .^ alpha) .* VSSimMatrix .* real((ISimMatrix .* QSimMatrix) .^ lambda) .* weight;
sim = sum(sum(SimMatrixC)) / sum(weight(:));
return;
%===================================
function VSMap = SDSP(image,sigmaF,omega0,sigmaD,sigmaC)
% ========================================================================
% SDSP algorithm for salient region detection from a given image.
% Copyright(c) 2013 Lin ZHANG, School of Software Engineering, Tongji
% University
% All Rights Reserved.
% ----------------------------------------------------------------------
% Permission to use, copy, or modify this software and its documentation
% for educational and research purposes only and without fee is here
% granted, provided that this copyright notice and the original authors'
% names appear on all copies and supporting documentation. This program
% shall not be used, rewritten, or adapted as the basis of a commercial
% software or hardware product without first obtaining permission of the
% authors. The authors make no representations about the suitability of
% this software for any purpose. It is provided "as is" without express
% or implied warranty.
%----------------------------------------------------------------------
%
% This is an implementation of the algorithm for calculating the
% SDSP (Saliency Detection by combining Simple Priors).
%
% Please refer to the following paper
%
% Lin Zhang, Zhongyi Gu, and Hongyu Li,"SDSP: a novel saliency detection
% method by combining simple priors", ICIP, 2013.
%
%----------------------------------------------------------------------
%
%Input : image: an uint8 RGB image with dynamic range [0, 255] for each
%channel
%
%Output: VSMap: the visual saliency map extracted by the SDSP algorithm.
%Data range for VSMap is [0, 255]. So, it can be regarded as a common
%gray-scale image.
%
%-----------------------------------------------------------------------
%convert the image into LAB color space
[oriRows, oriCols, junk] = size(image);
image = double(image);
dsImage(:,:,1) = imresize(image(:,:,1), [256, 256],'bilinear');
dsImage(:,:,2) = imresize(image(:,:,2), [256, 256],'bilinear');
dsImage(:,:,3) = imresize(image(:,:,3), [256, 256],'bilinear');
lab = RGB2Lab(dsImage);
LChannel = lab(:,:,1);
AChannel = lab(:,:,2);
BChannel = lab(:,:,3);
LFFT = fft2(double(LChannel));
AFFT = fft2(double(AChannel));
BFFT = fft2(double(BChannel));
[rows, cols, junk] = size(dsImage);
LG = logGabor(rows,cols,omega0,sigmaF);
FinalLResult = real(ifft2(LFFT.*LG));
FinalAResult = real(ifft2(AFFT.*LG));
FinalBResult = real(ifft2(BFFT.*LG));
SFMap = sqrt(FinalLResult.^2 + FinalAResult.^2 + FinalBResult.^2);
%the central areas will have a bias towards attention
coordinateMtx = zeros(rows, cols, 2);
coordinateMtx(:,:,1) = repmat((1:1:rows)', 1, cols);
coordinateMtx(:,:,2) = repmat(1:1:cols, rows, 1);
centerY = rows / 2;
centerX = cols / 2;
centerMtx(:,:,1) = ones(rows, cols) * centerY;
centerMtx(:,:,2) = ones(rows, cols) * centerX;
SDMap = exp(-sum((coordinateMtx - centerMtx).^2,3) / sigmaD^2);
%warm colors have a bias towards attention
maxA = max(AChannel(:));
minA = min(AChannel(:));
normalizedA = (AChannel - minA) / (maxA - minA);
maxB = max(BChannel(:));
minB = min(BChannel(:));
normalizedB = (BChannel - minB) / (maxB - minB);
labDistSquare = normalizedA.^2 + normalizedB.^2;
SCMap = 1 - exp(-labDistSquare / (sigmaC^2));
% VSMap = SFMap .* SDMap;
VSMap = SFMap .* SDMap .* SCMap;
VSMap = imresize(VSMap, [oriRows, oriCols],'bilinear');
VSMap = mat2gray(VSMap);
return;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function labImage = RGB2Lab(image)
image = double(image);
normalizedR = image(:,:,1) / 255;
normalizedG = image(:,:,2) / 255;
normalizedB = image(:,:,3) / 255;
RSmallerOrEqualto4045 = normalizedR <= 0.04045;
RGreaterThan4045 = 1 - RSmallerOrEqualto4045;
tmpR = (normalizedR / 12.92) .* RSmallerOrEqualto4045;
tmpR = tmpR + power((normalizedR + 0.055)/1.055,2.4) .* RGreaterThan4045;
GSmallerOrEqualto4045 = normalizedG <= 0.04045;
GGreaterThan4045 = 1 - GSmallerOrEqualto4045;
tmpG = (normalizedG / 12.92) .* GSmallerOrEqualto4045;
tmpG = tmpG + power((normalizedG + 0.055)/1.055,2.4) .* GGreaterThan4045;
BSmallerOrEqualto4045 = normalizedB <= 0.04045;
BGreaterThan4045 = 1 - BSmallerOrEqualto4045;
tmpB = (normalizedB / 12.92) .* BSmallerOrEqualto4045;
tmpB = tmpB + power((normalizedB + 0.055)/1.055,2.4) .* BGreaterThan4045;
X = tmpR*0.4124564 + tmpG*0.3575761 + tmpB*0.1804375;
Y = tmpR*0.2126729 + tmpG*0.7151522 + tmpB*0.0721750;
Z = tmpR*0.0193339 + tmpG*0.1191920 + tmpB*0.9503041;
epsilon = 0.008856; %actual CIE standard
kappa = 903.3; %actual CIE standard
Xr = 0.9642; %reference white D50
Yr = 1.0; %reference white
Zr = 0.8251; %reference white
xr = X/Xr;
yr = Y/Yr;
zr = Z/Zr;
xrGreaterThanEpsilon = xr > epsilon;
xrSmallerOrEqualtoEpsilon = 1 - xrGreaterThanEpsilon;
fx = power(xr, 1.0/3.0) .* xrGreaterThanEpsilon;
fx = fx + (kappa*xr + 16.0)/116.0 .* xrSmallerOrEqualtoEpsilon;
yrGreaterThanEpsilon = yr > epsilon;
yrSmallerOrEqualtoEpsilon = 1 - yrGreaterThanEpsilon;
fy = power(yr, 1.0/3.0) .* yrGreaterThanEpsilon;
fy = fy + (kappa*yr + 16.0)/116.0 .* yrSmallerOrEqualtoEpsilon;
zrGreaterThanEpsilon = zr > epsilon;
zrSmallerOrEqualtoEpsilon = 1 - zrGreaterThanEpsilon;
fz = power(zr, 1.0/3.0) .* zrGreaterThanEpsilon;
fz = fz + (kappa*zr + 16.0)/116.0 .* zrSmallerOrEqualtoEpsilon;
[rows,cols,junk] = size(image);
labImage = zeros(rows,cols,3);
labImage(:,:,1) = 116.0 * fy - 16.0;
labImage(:,:,2) = 500.0 * (fx - fy);
labImage(:,:,3) = 200.0 * (fy - fz);
return;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function LG = logGabor(rows,cols,omega0,sigmaF)
[u1, u2] = meshgrid(([1:cols]-(fix(cols/2)+1))/(cols-mod(cols,2)), ...
([1:rows]-(fix(rows/2)+1))/(rows-mod(rows,2)));
mask = ones(rows, cols);
for rowIndex = 1:rows
for colIndex = 1:cols
if u1(rowIndex, colIndex)^2 + u2(rowIndex, colIndex)^2 > 0.25
mask(rowIndex, colIndex) = 0;
end
end
end
u1 = u1 .* mask;
u2 = u2 .* mask;
u1 = ifftshift(u1);
u2 = ifftshift(u2);
radius = sqrt(u1.^2 + u2.^2);
radius(1,1) = 1;
LG = exp((-(log(radius/omega0)).^2) / (2 * (sigmaF^2)));
LG(1,1) = 0;
return;