Skip to content

Commit

Permalink
add
Browse files Browse the repository at this point in the history
  • Loading branch information
@csjunxu
csjunxu committed Jul 13, 2018
1 parent 3c3fc41 commit a7d578b
Show file tree
Hide file tree
Showing 22 changed files with 139 additions and 471 deletions.
83 changes: 83 additions & 0 deletions Demo_TWSC_RGB2YUV_CC2016.m
View file
@@ -0,0 +1,83 @@
%-------------------------------------------------------------------------------------------------------------
% This is an implementation of the TWSC algorithm for real-world image denoising
% by processing the YUV channels insead of RGB channels, it would improve
% TWSC by at 0.1dB on PSNR.
% Author: Jun Xu, csjunxu@comp.polyu.edu.hk / nankaimathxujun@gmail.com
% The Hong Kong Polytechnic University
%
% Please refer to the following paper if you find this code helps:
%
% @article{TWSC_ECCV2018,
% author = {Jun Xu and Lei Zhang and David Zhang},
% title = {A Trilateral Weighted Sparse Coding Scheme for Real-World Image Denoising},
% journal = {ECCV},
% year = {2018}
% }
%
% Please see the file License.txt for the license governing this code.
%-------------------------------------------------------------------------------------------------------------
clear;
GT_Original_image_dir = 'Real_ccnoise_denoised_Part/';
GT_fpath = fullfile(GT_Original_image_dir, '*mean.png');
TT_Original_image_dir = 'Real_ccnoise_denoised_Part/';
TT_fpath = fullfile(TT_Original_image_dir, '*real.png');
GT_im_dir = dir(GT_fpath);
TT_im_dir = dir(TT_fpath);
im_num = length(TT_im_dir);

method = 'TWSC_rgb2yuv';
dataset = 'cc';
write_MAT_dir = [dataset '_Results/'];
write_sRGB_dir = [write_MAT_dir method];
if ~isdir(write_sRGB_dir)
mkdir(write_sRGB_dir)
end

% Parameters
Par.ps = 6; % patch size
Par.step = 3; % the step of two neighbor patches
Par.win = 20; % size of window around the patch
Par.Outerloop = 7;
Par.Innerloop = 2;
Par.nlspini = 70;
Par.display = 0;
Par.delta = 0;
Par.nlspgap = 0;
Par.lambda1 = 0;
Par.lambda2 = 4.9;
Par.nlspini = 70;
% record all the results in each iteration
Par.PSNR = zeros(Par.Outerloop, im_num, 'double');
Par.SSIM = zeros(Par.Outerloop, im_num, 'double');
for i = 1 : im_num
Par.nlsp = Par.nlspini; % number of non-local patches
Par.image = i;
Par.nim = im2double(imread(fullfile(TT_Original_image_dir, TT_im_dir(i).name) ));
Par.I = im2double(imread(fullfile(GT_Original_image_dir, GT_im_dir(i).name)));
S = regexp(TT_im_dir(i).name, '\.', 'split');
IMname = S{1};
% initial PSNR and SSIM
fprintf('%s: \n', TT_im_dir(i).name);
fprintf('The initial PSNR = %2.4f, SSIM = %2.4f. \n', csnr( Par.nim*255, Par.I*255, 0, 0 ), cal_ssim( Par.nim*255, Par.I*255, 0, 0 ));

Par.nim = rgb2ycbcr(Par.nim);
[h,w,ch] = size(Par.nim);
% noise estimation
for c = 1:ch
Par.nSig(c) = NoiseEstimation(Par.nim(:, :, c)*255, Par.ps)/255;
end
[IMout, Par] = TWSC_Sigma_RW(Par);
IMout = ycbcr2rgb(IMout);
% calculate the PSNR
Par.PSNR(Par.Outerloop, Par.image) = csnr( IMout*255, Par.I*255, 0, 0 );
Par.SSIM(Par.Outerloop, Par.image) = cal_ssim( IMout*255, Par.I*255, 0, 0 );
%% output
imwrite(IMout, [write_sRGB_dir '/' method '_rgb2yuv_' dataset '_' IMname '.png']);
fprintf('%s : PSNR = %2.4f, SSIM = %2.4f \n',TT_im_dir(i).name, Par.PSNR(Par.Outerloop, Par.image),Par.SSIM(Par.Outerloop, Par.image) );
end
PSNR =Par.PSNR(end,:);
SSIM = Par.SSIM(end,:);
mPSNR=mean(PSNR,2);
mSSIM=mean(SSIM,2);
matname = sprintf([write_MAT_dir method '_rgb2yuv_' dataset '.mat']);
save(matname,'PSNR','SSIM','mPSNR','mSSIM');
275 changes: 53 additions & 222 deletions Demo_TWSC_RGB2YUV_DND2017.m
View file
@@ -1,245 +1,76 @@
%-------------------------------------------------------------------------------------------------------------
% This is an implementation of the TWSC algorithm for real-world image denoising
% by processing the YUV channels insead of RGB channels, it would improve
% TWSC by at 0.1dB on PSNR.
% Author: Jun Xu, csjunxu@comp.polyu.edu.hk / nankaimathxujun@gmail.com
% The Hong Kong Polytechnic University
%
% Please refer to the following paper if you find this code helps:
%
% @article{TWSC_ECCV2018,
% author = {Jun Xu and Lei Zhang and David Zhang},
% title = {A Trilateral Weighted Sparse Coding Scheme for Real-World Image Denoising},
% journal = {ECCV},
% year = {2018}
% }
%
% Please see the file License.txt for the license governing this code.
%-------------------------------------------------------------------------------------------------------------
clear;

Original_image_dir = 'C:\Users\csjunxu\Desktop\CVPR2018 Denoising\dnd_2017\images_srgb\';
fpath = fullfile(Original_image_dir, '*.mat');
im_dir = dir(fpath);
im_num = length(im_dir);
load 'C:\Users\csjunxu\Desktop\CVPR2018 Denoising\dnd_2017\info.mat';

method = 'TWSCrgb2yuv_y';
% write image directory
write_MAT_dir = ['C:/Users/csjunxu/Desktop/CVPR2018 Denoising/dnd_2017Results/'];
write_sRGB_dir = [write_MAT_dir method];
if ~isdir(write_sRGB_dir)
mkdir(write_sRGB_dir)
end
dataset = 'DND';
% Parameters
Par.ps = 6; % patch size
Par.step = 3; % the step of two neighbor patches
Par.win = 20; % size of window around the patch
Par.Outerloop = 8;
Par.Innerloop = 2;
Par.maxIter = 10;
Par.maxrho = 100;
Par.nlspini = 70;
Par.model = 1;
Par.display = 0;
Par.delta = 0;
Par.nlspgap = 0; %10

for mu = [1.1]
Par.mu = mu;
for rho = [0.5]
Par.rho = rho;
for lambda1 = [0]
Par.lambda1 = lambda1;
for lambda2 = [3]
Par.lambda2 = lambda2;
for nlspini = [70]
Par.nlspini = nlspini;
% set Parameters
% record all the results in each iteration
Par.PSNR = zeros(Par.Outerloop, im_num*20, 'double');
Par.SSIM = zeros(Par.Outerloop, im_num*20, 'double');
CCPSNR = [];
CCSSIM = [];
for i = 1 :im_num
Par.image = i;
load(fullfile(Original_image_dir, im_dir(i).name));
S = regexp(im_dir(i).name, '\.', 'split');
[h,w,ch] = size(InoisySRGB);
for j = 1:size(info(1).boundingboxes,1)
Par.nlsp = Par.nlspini; % number of non-local patches
IMinname = [S{1} '_' num2str(j)];
bb = info(i).boundingboxes(j,:);
Par.nim = InoisySRGB(bb(1):bb(3), bb(2):bb(4),:);
Par.I = Par.nim;
Par.nim = rgb2ycbcr(Par.nim);
Par.nimcb = Par.nim(:,:,2);
Par.nimcr = Par.nim(:,:,3);
Par.nim = Par.nim(:,:,1);
[h,w,ch] = size(Par.nim);
% noise estimation
for c = 1:ch
% Par.nSig0(c) = NoiseLevel(Par.nim(:, :, c));
Par.nSig(c) = NoiseEstimation(Par.nim(:, :, c)*255, Par.ps)/255;
end
[IMout, Par] = TWSC_Sigma_WAR(Par);
IMout(:,:,2) = Par.nimcb;
IMout(:,:,3) = Par.nimcr;
IMout = ycbcr2rgb(IMout);
% initial PSNR and SSIM
fprintf('%s: \n', IMinname);

% calculate the PSNR
Par.PSNR(Par.Outerloop, Par.image) = csnr( IMout*255, Par.I*255, 0, 0 );
Par.SSIM(Par.Outerloop, Par.image) = cal_ssim( IMout*255, Par.I*255, 0, 0 );
%% output
imwrite(IMout, [write_sRGB_dir '/' method '_' dataset '_l1_' num2str(lambda1) '_l2_' num2str(lambda2) '_N' num2str(nlspini) '_' IMinname '.png']);
end
end
end
end
end
end
end

Original_image_dir = 'C:\Users\csjunxu\Desktop\CVPR2018 Denoising\dnd_2017\images_srgb\';
fpath = fullfile(Original_image_dir, '*.mat');
im_dir = dir(fpath);
im_num = length(im_dir);
load 'C:\Users\csjunxu\Desktop\CVPR2018 Denoising\dnd_2017\info.mat';

method = 'TWSCrgb2yuv_y_u_v';
% write image directory
write_MAT_dir = ['C:/Users/csjunxu/Desktop/CVPR2018 Denoising/dnd_2017Results/'];
write_sRGB_dir = [write_MAT_dir method];
if ~isdir(write_sRGB_dir)
mkdir(write_sRGB_dir)
end
dataset = 'DND';
% Parameters
Par.ps = 6; % patch size
Par.step = 3; % the step of two neighbor patches
Par.win = 20; % size of window around the patch
Par.Outerloop = 8;
Par.Innerloop = 2;
Par.maxIter = 10;
Par.maxrho = 100;
Par.nlspini = 70;
Par.model = 1;
Par.display = 0;
Par.delta = 0;
Par.nlspgap = 0; %10

for mu = [1.1]
Par.mu = mu;
for rho = [0.5]
Par.rho = rho;
for lambda1 = [0]
Par.lambda1 = lambda1;
for lambda2 = [3]
Par.lambda2 = lambda2;
for nlspini = [70]
Par.nlspini = nlspini;
% set Parameters
% record all the results in each iteration
Par.PSNR = zeros(Par.Outerloop, im_num*20, 'double');
Par.SSIM = zeros(Par.Outerloop, im_num*20, 'double');
CCPSNR = [];
CCSSIM = [];
for i = 1 :im_num
Par.image = i;
load(fullfile(Original_image_dir, im_dir(i).name));
S = regexp(im_dir(i).name, '\.', 'split');
[h,w,ch] = size(InoisySRGB);
for j = 1:size(info(1).boundingboxes,1)
Par.nlsp = Par.nlspini; % number of non-local patches
IMinname = [S{1} '_' num2str(j)];
bb = info(i).boundingboxes(j,:);
Par.nim = InoisySRGB(bb(1):bb(3), bb(2):bb(4),:);
Par.I = Par.nim;
Par.nimycbcr = rgb2ycbcr(Par.nim);
[h,w,ch] = size(Par.nim);
% noise estimation
for c = 1:ch
Par.nim = Par.nimycbcr(:, :, c);
Par.nSig(c) = NoiseEstimation(Par.nim*255, Par.ps)/255;
% denoising
[IMout(:,:,c), Par] = TWSC_Sigma_WAR(Par);
end
IMout = ycbcr2rgb(IMout);
% calculate the PSNR
Par.PSNR(Par.Outerloop, Par.image) = csnr( IMout*255, Par.I*255, 0, 0 );
Par.SSIM(Par.Outerloop, Par.image) = cal_ssim( IMout*255, Par.I*255, 0, 0 );
%% output
imwrite(IMout, [write_sRGB_dir '/' method '_' dataset '_l1_' num2str(lambda1) '_l2_' num2str(lambda2) '_N' num2str(nlspini) '_' IMinname '.png']);
end
end
end
end
end
end
end

clear;

Original_image_dir = 'C:\Users\csjunxu\Desktop\CVPR2018 Denoising\dnd_2017\images_srgb\';
Original_image_dir = 'dnd_2017/images_srgb/';
fpath = fullfile(Original_image_dir, '*.mat');
im_dir = dir(fpath);
im_num = length(im_dir);
load 'C:\Users\csjunxu\Desktop\CVPR2018 Denoising\dnd_2017\info.mat';
load 'dnd_2017/info.mat';

method = 'TWSCrgb2yuv';
method = 'TWSC_rgb2yuv';
dataset = 'dnd';
% write image directory
write_MAT_dir = ['C:/Users/csjunxu/Desktop/CVPR2018 Denoising/dnd_2017Results/'];
write_MAT_dir = [dataset '_Results/'];
write_sRGB_dir = [write_MAT_dir method];
if ~isdir(write_sRGB_dir)
mkdir(write_sRGB_dir)
end
dataset = 'DND';
% Parameters
Par.ps = 6; % patch size
Par.step = 3; % the step of two neighbor patches
Par.win = 20; % size of window around the patch
Par.Outerloop = 8;
Par.Innerloop = 2;
Par.maxIter = 10;
Par.maxrho = 100;
Par.nlspini = 70;
Par.model = 1;
Par.display = 0;
Par.delta = 0;
Par.nlspgap = 0; %10

for mu = [1.1]
Par.mu = mu;
for rho = [0.5]
Par.rho = rho;
for lambda1 = [0]
Par.lambda1 = lambda1;
for lambda2 = [3]
Par.lambda2 = lambda2;
for nlspini = [70]
Par.nlspini = nlspini;
% set Parameters
% record all the results in each iteration
Par.PSNR = zeros(Par.Outerloop, im_num*20, 'double');
Par.SSIM = zeros(Par.Outerloop, im_num*20, 'double');
CCPSNR = [];
CCSSIM = [];
for i = 1 :im_num
Par.image = i;
load(fullfile(Original_image_dir, im_dir(i).name));
S = regexp(im_dir(i).name, '\.', 'split');
[h,w,ch] = size(InoisySRGB);
for j = 1:size(info(1).boundingboxes,1)
Par.nlsp = Par.nlspini; % number of non-local patches
IMinname = [S{1} '_' num2str(j)];
bb = info(i).boundingboxes(j,:);
Par.nim = InoisySRGB(bb(1):bb(3), bb(2):bb(4),:);
Par.I = Par.nim;
Par.nim = rgb2ycbcr(Par.nim);
[h,w,ch] = size(Par.nim);
% noise estimation
for c = 1:ch
% Par.nSig0(c) = NoiseLevel(Par.nim(:, :, c));
Par.nSig(c) = NoiseEstimation(Par.nim(:, :, c)*255, Par.ps)/255;
end
[IMout, Par] = TWSC_Sigma_WAR(Par);
IMout = ycbcr2rgb(IMout);
% initial PSNR and SSIM
fprintf('%s: \n', IMinname);
% calculate the PSNR
Par.PSNR(Par.Outerloop, Par.image) = csnr( IMout*255, Par.I*255, 0, 0 );
Par.SSIM(Par.Outerloop, Par.image) = cal_ssim( IMout*255, Par.I*255, 0, 0 );
%% output
imwrite(IMout, [write_sRGB_dir '/' method '_' dataset '_l1_' num2str(lambda1) '_l2_' num2str(lambda2) '_N' num2str(nlspini) '_' IMinname '.png']);
end
end
end
end
Par.lambda1 = 0;
Par.lambda2 = 3;
% record all the results in each iteration
Par.PSNR = zeros(Par.Outerloop, im_num*20, 'double');
Par.SSIM = zeros(Par.Outerloop, im_num*20, 'double');
for i = 1 :im_num
Par.image = i;
load(fullfile(Original_image_dir, im_dir(i).name));
S = regexp(im_dir(i).name, '\.', 'split');
[h,w,ch] = size(InoisySRGB);
for j = 1:size(info(1).boundingboxes,1)
Par.nlsp = Par.nlspini; % number of non-local patches
IMinname = [S{1} '_' num2str(j)];
bb = info(i).boundingboxes(j,:);
Par.nim = InoisySRGB(bb(1):bb(3), bb(2):bb(4),:);
Par.I = Par.nim;
Par.nim = rgb2ycbcr(Par.nim);
[h,w,ch] = size(Par.nim);
% noise estimation
for c = 1:ch
Par.nSig(c) = NoiseEstimation(Par.nim(:, :, c)*255, Par.ps)/255;
end
[IMout, Par] = TWSC_Sigma_RW(Par);
IMout = ycbcr2rgb(IMout);
% initial PSNR and SSIM
fprintf('%s: \n', IMinname);
% calculate the PSNR
Par.PSNR(Par.Outerloop, Par.image) = csnr( IMout*255, Par.I*255, 0, 0 );
Par.SSIM(Par.Outerloop, Par.image) = cal_ssim( IMout*255, Par.I*255, 0, 0 );
%% output
imwrite(IMout, [write_sRGB_dir '/' method '_' dataset '_' IMinname '.png']);
end
end

0 comments on commit a7d578b

Please sign in to comment.