-
Notifications
You must be signed in to change notification settings - Fork 174
/
Demo_SISR_direct_downsampler_YCbCr.m
212 lines (178 loc) · 8.49 KB
/
Demo_SISR_direct_downsampler_YCbCr.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
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Demo of IRCNN for image super-resolution where the latent HR image x is blurred and then downsampled to get the LR image y
% (y can be corrupted by additive Gaussian noise of level Isigma).
%
% The details of this degradation can be found by the following paper.
% [1] S. H. Chan, X. Wang, and O. A. Elgendy "Plug-and-Play ADMM for image restoration: Fixed point convergence and applications", IEEE Transactions on Computational Imaging, 2016.
%
% The objective function is given by min_x 1/(Isigma^2)||x*k_{direct downsampler with scale factor sf}-y||^2 + lamda Phi(x)
%
% k -- blur kernel, not limited to Gaussian blur
% direct downsampler -- implemented by matlab function "downsample",
% sf -- scale factor, 2,3,4,...
% Isigma -- estimated noise level of y, should be slightly larger than the true one.
%
% @inproceedings{zhang2017learning,
% title={Learning Deep CNN Denoiser Prior for Image Restoration},
% author={Zhang, Kai and Zuo, Wangmeng and Gu, Shuhang and Zhang, Lei},
% booktitle={IEEE Conference on Computer Vision and Pattern Recognition},
% year={2017}
% }
% If you have any question, please feel free to contact with me.
% Kai Zhang (e-mail: cskaizhang@gmail.com)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% clear; clc;
addpath('utilities');
imageSets = {'Set5','Set14','BSD100','Urban100'}; % testing dataset
%%% setting
setTest = imageSets([1]); % select the dataset
showResult = 1;
pauseTime = 0;
useGPU = 1; % 1 or 0, true or false
folderTest = 'testsets';
folderResult= 'results';
taskTestCur = 'SISR';
if ~exist(folderResult,'file')
mkdir(folderResult);
end
%% parameter setting of HQS
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%Important!%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
sf = 3; % scale factor
Isigma = 0.5/255; % noise level of y (Note: for RGB images, y is the Y channel of YCbCr space.) from [0.5, 50]/255
Isigma = max(Isigma,0.1/255);
Msigma = sf; % noise level of last denoiser, from [1,15]
% blur kernel k, not limited to Gaussian blur
kernelsigma = 1.6; % width (sigma) of the Gaussian blur kernel
% from [0.6 2.4], e.g., sf = 2, kernelsigma = 1; sf = 3, kernelsigma = 1.6; sf = 4, kernelsigma = 2;
k = fspecial('gaussian',7,kernelsigma);
%k = fspecial('motion',20,45); % You can try this motion blur kernel ^_^
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% default parameter setting of HQS
totalIter = 30; % default 30
modelSigmaS = logspace(log10(49),log10(Msigma),totalIter); % default 49, or 3*sf
ns = min(25,max(ceil(modelSigmaS/2),1));
ns = [ns(1)-1,ns];
lamda = (Isigma^2)/3; % default 3, ****** from {1 2 3 4} ******
%% load denoiser model
folderModel = 'models';
load(fullfile(folderModel,'modelgray.mat'));
%% do SISR
for n_set = 1 : numel(setTest)
% read images
setTestCur = cell2mat(setTest(n_set));
disp('--------------------------------------------');
disp(['----',setTestCur,'-----Super-Resolution-----']);
disp('--------------------------------------------');
folderTestCur = fullfile(folderTest,setTestCur);
ext = {'*.jpg','*.png','*.bmp'};
filepaths = [];
for i = 1 : length(ext)
filepaths = cat(1,filepaths,dir(fullfile(folderTestCur, ext{i})));
end
eval(['PSNR_',taskTestCur,'_',setTestCur,'_x',num2str(sf),' = zeros(length(filepaths),1);']);
eval(['PSNRC_',taskTestCur,'_',setTestCur,'_x',num2str(sf),' = zeros(length(filepaths),1);']);
% folder to store results
folderResultCur = fullfile(folderResult, ['SISR_YCbCr_direct_downsample_',setTestCur,'_x',num2str(sf)]);
if ~exist(folderResultCur,'file')
mkdir(folderResultCur);
end
for i = 1 : length(filepaths)
HR = imread(fullfile(folderTestCur,filepaths(i).name));
[~,imageName,ext] = fileparts(filepaths(i).name);
HR = modcrop(HR, sf);
% label_RGB (uint8)
label_RGB = HR;
chanel = size(HR,3);
%%%%%%%%%%%%%%%%%%%%%% image degradation %%%%%%%%%%%%%%%%%%%%
% LR (uint8), get the LR image
blur_HR = imfilter(HR,k,'circular'); % blurred
LR = downsample2(blur_HR, sf); % downsampled
LR = uint8(LR);
if chanel == 3
% label (single)
HR_ycc = single(rgb2ycbcr(im2double(HR)));
label = HR_ycc(:,:,1);
LR_ycc = single(rgb2ycbcr(im2double(LR)));
LRY = LR_ycc(:,:,1);
% input (single)
HR_bic = imresize(im2double(LR),sf,'bicubic');
LR_bic_ycc = rgb2ycbcr(HR_bic);
input = im2single(LR_bic_ycc(:,:,1));
% input_RGB (uint8)
input_RGB = im2uint8(HR_bic);
else
% label (single)
label = im2single(HR);
LRY = im2single(LR);
HR_bic = imresize(LRY,sf,'bicubic');
% input (single)
input = im2single(HR_bic);
% input_RGB (uint8)
input_RGB = HR_bic;
end
y = im2single(LRY);
[rows_in,cols_in] = size(y);
rows = rows_in*sf;
cols = cols_in*sf;
[G,Gt] = defGGt(k,sf);
GGt = constructGGt(k,sf,rows,cols);
Gty = Gt(y);
if useGPU
input = gpuArray(input);
LRY = gpuArray(LRY);
end
output = input;
tic;
for itern = 1:totalIter
% step 1, closed-form solution, see Chan et al. [1] for details
rho = lamda*255^2/(modelSigmaS(itern)^2);
rhs = Gty + rho*output;
output = (rhs - Gt(real(ifft2(fft2(G(rhs))./(GGt + rho)))))/rho;
% load denoiser
if ns(itern+1)~=ns(itern)
[net] = loadmodel(modelSigmaS(itern),CNNdenoiser);
net = vl_simplenn_tidy(net);
if useGPU
net = vl_simplenn_move(net, 'gpu');
end
end
% step 2, perform denoising
res = vl_simplenn(net, output,[],[],'conserveMemory',true,'mode','test');
im = res(end).x; % residual image
output = output - im;
% imshow(output)
% drawnow;
% pause(1)
end
if useGPU
output = gather(output);
end
toc;
if chanel == 3
% output_RGB (uint8)
LR_bic_ycc(:,:,1) = double(output);
output_RGB = im2uint8(ycbcr2rgb(LR_bic_ycc));
else
% output_RGB (uint8)
output_RGB = im2uint8(output);
end
[PSNR_Cur,SSIM_Cur] = Cal_PSNRSSIM(label*255,output*255,ceil(sf),ceil(sf)); % calculate PSNR and SSIM on Y channel of YCbCr space
[PSNRC_Cur,SSIM_Cur_RGB] = Cal_PSNRSSIM(label_RGB,output_RGB,ceil(sf),ceil(sf)); % calculate PSNR and SSIM on R,G,B channels
disp(['Single Image Super-Resolution ',num2str(PSNR_Cur,'%2.2f'),'dB',' ',filepaths(i).name]);
eval(['PSNR_',taskTestCur,'_',setTestCur,'_x',num2str(sf),'(',num2str(i),') = PSNR_Cur;']);
eval(['PSNRC_',taskTestCur,'_',setTestCur,'_x',num2str(sf),'(',num2str(i),') = PSNRC_Cur;']);
if showResult
imshow(cat(2,input_RGB,output_RGB,label_RGB));
drawnow;
title(['Single Image Super-Resolution ',filepaths(i).name,' ',num2str(PSNR_Cur,'%2.2f'),'dB'],'FontSize',12)
pause(pauseTime)
%pause()
imwrite(output_RGB,fullfile(folderResultCur,[imageName,'_x',num2str(sf),'.png']));
end
end
disp(['Average PSNR on Y is ',num2str(mean(eval(['PSNR_',taskTestCur,'_',setTestCur,'_x',num2str(sf)])),'%2.2f'),'dB']);
disp(['Average PSNR on RGB is ',num2str(mean(eval(['PSNRC_',taskTestCur,'_',setTestCur,'_x',num2str(sf)])),'%2.2f'),'dB']);
% save PSNR and SSIM metrics
save(fullfile(folderResultCur,['PSNR_',taskTestCur,'_',setTestCur,'_x',num2str(sf),'.mat']),['PSNR_',taskTestCur,'_',setTestCur,'_x',num2str(sf)])
save(fullfile(folderResultCur,['PSNRC_',taskTestCur,'_',setTestCur,'_x',num2str(sf),'.mat']),['PSNRC_',taskTestCur,'_',setTestCur,'_x',num2str(sf)])
end