Mercurial > hg > smallbox

annotate examples/Image Denoising/SMALL_ImgDenoise_DL_test_KSVDvsRLSDLA.m @ 78:f69ae88b8be5

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
added Rice Wavelet Toolbox with my modification, so it can be compiled on newer systems.
author Ivan Damnjanovic lnx <ivan.damnjanovic@eecs.qmul.ac.uk>
date Fri, 25 Mar 2011 15:27:33 +0000
parents 55faa9b5d1ac
children 4302a91e6033
rev   line source
idamnjanovic@42 1 %% DICTIONARY LEARNING FOR IMAGE DENOISING
idamnjanovic@42 2 % This file contains an example of how SMALLbox can be used to test different
idamnjanovic@42 3 % dictionary learning techniques in Image Denoising problem.
idamnjanovic@42 4 % It calls generateImageDenoiseProblem that will let you to choose image,
idamnjanovic@42 5 % add noise and use noisy image to generate training set for dictionary
idamnjanovic@42 6 % learning.
idamnjanovic@42 7 % Three dictionary learning techniques were compared:
idamnjanovic@42 8 % - KSVD - M. Elad, R. Rubinstein, and M. Zibulevsky, "Efficient
idamnjanovic@42 9 % Implementation of the K-SVD Algorithm using Batch Orthogonal
idamnjanovic@42 10 % Matching Pursuit", Technical Report - CS, Technion, April 2008.
idamnjanovic@42 11 % - KSVDS - R. Rubinstein, M. Zibulevsky, and M. Elad, "Learning Sparse
idamnjanovic@42 12 % Dictionaries for Sparse Signal Approximation", Technical
idamnjanovic@42 13 % Report - CS, Technion, June 2009.
idamnjanovic@42 14 % - SPAMS - J. Mairal, F. Bach, J. Ponce and G. Sapiro. Online
idamnjanovic@42 15 % Dictionary Learning for Sparse Coding. International
idamnjanovic@42 16 % Conference on Machine Learning,Montreal, Canada, 2009
idamnjanovic@42 17 %
idamnjanovic@42 18 %
idamnjanovic@42 19 % Ivan Damnjanovic 2010
idamnjanovic@42 20 %%
idamnjanovic@42 21
idamnjanovic@42 22
idamnjanovic@42 23
idamnjanovic@42 24 % If you want to load the image outside of generateImageDenoiseProblem
idamnjanovic@42 25 % function uncomment following lines. This can be useful if you want to
idamnjanovic@42 26 % denoise more then one image for example.
idamnjanovic@42 27 clear;
idamnjanovic@42 28 TMPpath=pwd;
idamnjanovic@42 29 FS=filesep;
idamnjanovic@42 30 [pathstr1, name, ext, versn] = fileparts(which('SMALLboxSetup.m'));
idamnjanovic@42 31 cd([pathstr1,FS,'data',FS,'images']);
idamnjanovic@42 32 load('test_image.mat');
ivan@78 33 cd(TMPpath);
idamnjanovic@42 34 % [filename,pathname] = uigetfile({'*.png;'},'Select a file containin pre-calculated notes');
idamnjanovic@42 35 % [pathstr, name, ext, versn] = fileparts(filename);
idamnjanovic@42 36 % test_image = imread(filename);
idamnjanovic@42 37 % test_image = double(test_image);
idamnjanovic@42 38 % cd(TMPpath);
idamnjanovic@42 39 % SMALL.Problem.name=name;
idamnjanovic@42 40
idamnjanovic@42 41 noise_level=[10 20 25 50 100];
idamnjanovic@42 42 % Defining Image Denoising Problem as Dictionary Learning
idamnjanovic@42 43 % Problem. As an input we set the number of training patches.
idamnjanovic@42 44 for noise_ind=1:1
ivan@78 45 for im_num=2:2
idamnjanovic@65 46 SMALL.Problem = generateImageDenoiseProblem(test_image(im_num).i, 40000, '',256, noise_level(noise_ind));
ivan@78 47 SMALL.Problem.name=int2str(im_num);
idamnjanovic@42 48
idamnjanovic@42 49 results(noise_ind,im_num).noisy_psnr=SMALL.Problem.noisy_psnr;
idamnjanovic@42 50
idamnjanovic@42 51 %%
idamnjanovic@42 52 % Use KSVD Dictionary Learning Algorithm to Learn overcomplete dictionary
idamnjanovic@42 53
idamnjanovic@42 54 % Initialising Dictionary structure
idamnjanovic@42 55 % Setting Dictionary structure fields (toolbox, name, param, D and time)
idamnjanovic@42 56 % to zero values
idamnjanovic@42 57
idamnjanovic@42 58 SMALL.DL(1)=SMALL_init_DL();
idamnjanovic@42 59
idamnjanovic@42 60 % Defining the parameters needed for dictionary learning
idamnjanovic@42 61
idamnjanovic@42 62 SMALL.DL(1).toolbox = 'KSVD';
idamnjanovic@42 63 SMALL.DL(1).name = 'ksvd';
idamnjanovic@42 64
idamnjanovic@42 65 % Defining the parameters for KSVD
idamnjanovic@42 66 % In this example we are learning 256 atoms in 20 iterations, so that
idamnjanovic@42 67 % every patch in the training set can be represented with target error in
idamnjanovic@42 68 % L2-norm (EData)
idamnjanovic@42 69 % Type help ksvd in MATLAB prompt for more options.
idamnjanovic@42 70
idamnjanovic@42 71 Edata=sqrt(prod(SMALL.Problem.blocksize)) * SMALL.Problem.sigma * SMALL.Problem.gain;
idamnjanovic@42 72 maxatoms = floor(prod(SMALL.Problem.blocksize)/2);
idamnjanovic@42 73 SMALL.DL(1).param=struct(...
idamnjanovic@42 74 'Edata', Edata,...
idamnjanovic@42 75 'initdict', SMALL.Problem.initdict,...
idamnjanovic@42 76 'dictsize', SMALL.Problem.p,...
idamnjanovic@42 77 'exact', 1, ...
idamnjanovic@42 78 'iternum', 20,...
idamnjanovic@42 79 'memusage', 'high');
idamnjanovic@42 80
idamnjanovic@42 81 % Learn the dictionary
idamnjanovic@42 82
idamnjanovic@42 83 SMALL.DL(1) = SMALL_learn(SMALL.Problem, SMALL.DL(1));
idamnjanovic@42 84
idamnjanovic@42 85 % Set SMALL.Problem.A dictionary
idamnjanovic@42 86 % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@42 87 % only with Problem structure, ie no direct communication between DL and
idamnjanovic@42 88 % solver structures)
idamnjanovic@42 89
idamnjanovic@42 90 SMALL.Problem.A = SMALL.DL(1).D;
idamnjanovic@42 91 SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem);
idamnjanovic@42 92
idamnjanovic@42 93 %%
idamnjanovic@42 94 % Initialising solver structure
idamnjanovic@42 95 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 96 % reconstructed and time) to zero values
idamnjanovic@42 97
idamnjanovic@42 98 SMALL.solver(1)=SMALL_init_solver;
idamnjanovic@42 99
idamnjanovic@42 100 % Defining the parameters needed for image denoising
idamnjanovic@42 101
idamnjanovic@42 102 SMALL.solver(1).toolbox='ompbox';
idamnjanovic@42 103 SMALL.solver(1).name='omp2';
idamnjanovic@42 104 SMALL.solver(1).param=struct(...
idamnjanovic@42 105 'epsilon',Edata,...
idamnjanovic@42 106 'maxatoms', maxatoms);
idamnjanovic@42 107
idamnjanovic@42 108 % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 109 % but backward compatible with KSVD definition of denoising
idamnjanovic@42 110
idamnjanovic@42 111 SMALL.solver(1)=SMALL_solve(SMALL.Problem, SMALL.solver(1));
idamnjanovic@42 112 SMALL.solver(1).reconstructed.psnr
idamnjanovic@42 113 %%
idamnjanovic@42 114 % Use KSVDS Dictionary Learning Algorithm to denoise image
idamnjanovic@42 115
idamnjanovic@42 116 % Initialising solver structure
idamnjanovic@42 117 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 118 % reconstructed and time) to zero values
idamnjanovic@42 119 %
idamnjanovic@42 120 % SMALL.DL(2)=SMALL_init_DL();
idamnjanovic@42 121 %
idamnjanovic@42 122 % % Defining the parameters needed for dictionary learning
idamnjanovic@42 123 %
idamnjanovic@42 124 % SMALL.DL(2).toolbox = 'KSVDS';
idamnjanovic@42 125 % SMALL.DL(2).name = 'ksvds';
idamnjanovic@42 126 %
idamnjanovic@42 127 % % Defining the parameters for KSVDS
idamnjanovic@42 128 % % In this example we are learning 256 atoms in 20 iterations, so that
idamnjanovic@42 129 % % every patch in the training set can be represented with target error in
idamnjanovic@42 130 % % L2-norm (EDataS). We also impose "double sparsity" - dictionary itself
idamnjanovic@42 131 % % has to be sparse in the given base dictionary (Tdict - number of
idamnjanovic@42 132 % % nonzero elements per atom).
idamnjanovic@42 133 % % Type help ksvds in MATLAB prompt for more options.
idamnjanovic@42 134 %
idamnjanovic@42 135 %
idamnjanovic@42 136 % SMALL.DL(2).param=struct(...
idamnjanovic@42 137 % 'Edata', Edata, ...
idamnjanovic@42 138 % 'Tdict', 6,...
idamnjanovic@42 139 % 'stepsize', 1,...
idamnjanovic@42 140 % 'dictsize', SMALL.Problem.p,...
idamnjanovic@42 141 % 'iternum', 20,...
idamnjanovic@42 142 % 'memusage', 'high');
idamnjanovic@42 143 % SMALL.DL(2).param.initA = speye(SMALL.Problem.p);
idamnjanovic@42 144 % SMALL.DL(2).param.basedict{1} = odctdict(8,16);
idamnjanovic@42 145 % SMALL.DL(2).param.basedict{2} = odctdict(8,16);
idamnjanovic@42 146 %
idamnjanovic@42 147 % % Learn the dictionary
idamnjanovic@42 148 %
idamnjanovic@42 149 % SMALL.DL(2) = SMALL_learn(SMALL.Problem, SMALL.DL(2));
idamnjanovic@42 150
idamnjanovic@42 151 % Set SMALL.Problem.A dictionary and SMALL.Problem.basedictionary
idamnjanovic@42 152 % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@42 153 % only with Problem structure, ie no direct communication between DL and
idamnjanovic@42 154 % solver structures)
idamnjanovic@42 155
idamnjanovic@42 156 SMALL.Problem.A = SMALL.Problem.initdict;
idamnjanovic@42 157 % SMALL.Problem.basedict{1} = SMALL.DL(2).param.basedict{1};
idamnjanovic@42 158 % SMALL.Problem.basedict{2} = SMALL.DL(2).param.basedict{2};
idamnjanovic@42 159 SMALL.DL(2).D=SMALL.Problem.initdict;
idamnjanovic@42 160 SparseDict=0;
idamnjanovic@42 161 SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem, SparseDict);
idamnjanovic@42 162
idamnjanovic@42 163 %%
idamnjanovic@42 164 % Initialising solver structure
idamnjanovic@42 165 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 166 % reconstructed and time) to zero values
idamnjanovic@42 167
idamnjanovic@42 168 SMALL.solver(2)=SMALL_init_solver;
idamnjanovic@42 169
idamnjanovic@42 170 % Defining the parameters needed for image denoising
idamnjanovic@42 171
idamnjanovic@42 172 SMALL.solver(2).toolbox='ompbox';
idamnjanovic@42 173 SMALL.solver(2).name='omp2';
idamnjanovic@42 174 SMALL.solver(2).param=struct(...
idamnjanovic@42 175 'epsilon',Edata,...
idamnjanovic@42 176 'maxatoms', maxatoms);
idamnjanovic@42 177
idamnjanovic@42 178 % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 179 % but backward compatible with KSVD definition of denoising
idamnjanovic@42 180 % Pay attention that since implicit base dictionary is used, denoising
idamnjanovic@42 181 % can be much faster then using explicit dictionary in KSVD example.
idamnjanovic@42 182
idamnjanovic@42 183 SMALL.solver(2)=SMALL_solve(SMALL.Problem, SMALL.solver(2));
idamnjanovic@42 184 %%
idamnjanovic@42 185
idamnjanovic@42 186 for i =1:1
idamnjanovic@42 187
idamnjanovic@42 188 X=SMALL.Problem.b1;
idamnjanovic@42 189 X_norm=sqrt(sum(X.^2, 1));
idamnjanovic@42 190 [X_norm_sort, p]=sort(X_norm);
idamnjanovic@42 191 p1=p(X_norm_sort>Edata);
idamnjanovic@42 192 if size(p1,2)>140000
idamnjanovic@42 193 p2 = randperm(size(p1,2));
idamnjanovic@42 194 p2=sort(p2(1:40000));
idamnjanovic@42 195 size(p2,2)
idamnjanovic@42 196 SMALL.Problem.b=X(:,p1(p2));
idamnjanovic@42 197 else
idamnjanovic@42 198 size(p1,2)
idamnjanovic@42 199 SMALL.Problem.b=X(:,p1);
idamnjanovic@42 200
idamnjanovic@42 201 end
idamnjanovic@42 202
idamnjanovic@42 203 lambda=0.9998
idamnjanovic@42 204
idamnjanovic@42 205 % Use Recursive Least Squares
idamnjanovic@42 206 % to Learn overcomplete dictionary
idamnjanovic@42 207
idamnjanovic@42 208 % Initialising Dictionary structure
idamnjanovic@42 209 % Setting Dictionary structure fields (toolbox, name, param, D and time)
idamnjanovic@42 210 % to zero values
idamnjanovic@42 211
idamnjanovic@42 212 SMALL.DL(3)=SMALL_init_DL();
idamnjanovic@42 213
idamnjanovic@42 214 % Defining fields needed for dictionary learning
idamnjanovic@42 215
idamnjanovic@42 216 SMALL.DL(3).toolbox = 'SMALL';
idamnjanovic@42 217 SMALL.DL(3).name = 'SMALL_rlsdla';
idamnjanovic@42 218 SMALL.DL(3).param=struct(...
idamnjanovic@42 219 'Edata', Edata,...
idamnjanovic@42 220 'initdict', SMALL.Problem.initdict,...
idamnjanovic@42 221 'dictsize', SMALL.Problem.p,...
idamnjanovic@42 222 'forgettingMode', 'FIX',...
idamnjanovic@42 223 'forgettingFactor', lambda);
idamnjanovic@42 224
idamnjanovic@42 225 % % Type 'help mexTrainDL in MATLAB prompt for explanation of parameters.
idamnjanovic@42 226 %
idamnjanovic@42 227 % SMALL.DL(3).param=struct(...
idamnjanovic@42 228 % 'D', SMALL.Problem.initdict,...
idamnjanovic@42 229 % 'K', SMALL.Problem.p,...
idamnjanovic@42 230 % 'lambda', 2,...
idamnjanovic@42 231 % 'iter', 200,...
idamnjanovic@42 232 % 'mode', 3, ...
idamnjanovic@42 233 % 'modeD', 0);
idamnjanovic@42 234
idamnjanovic@42 235 % Learn the dictionary
idamnjanovic@42 236
idamnjanovic@42 237 SMALL.DL(3) = SMALL_learn(SMALL.Problem, SMALL.DL(3));
idamnjanovic@42 238 %SMALL.DL(3).D(:,1)=SMALL.DL(1).D(:,1);
idamnjanovic@42 239 %
idamnjanovic@42 240 % % Set SMALL.Problem.A dictionary
idamnjanovic@42 241 % % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@42 242 % % only with Problem structure, ie no direct communication between DL and
idamnjanovic@42 243 % % solver structures)
idamnjanovic@42 244 %
idamnjanovic@42 245 %
idamnjanovic@42 246 %
idamnjanovic@42 247 % %%
idamnjanovic@42 248 % % Initialising solver structure
idamnjanovic@42 249 % % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 250 % % reconstructed and time) to zero values
idamnjanovic@42 251 % SMALL.Problem.A = SMALL.DL(1).D;
idamnjanovic@42 252 % SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem);
idamnjanovic@42 253 % maxatoms=5;
idamnjanovic@42 254 % SMALL.solver(3)=SMALL_init_solver;
idamnjanovic@42 255 %
idamnjanovic@42 256 % % Defining the parameters needed for denoising
idamnjanovic@42 257 %
idamnjanovic@42 258 % % SMALL.solver(3).toolbox='SPAMS';
idamnjanovic@42 259 % % SMALL.solver(3).name='mexLasso';
idamnjanovic@42 260 % % SMALL.solver(3).param=struct(...
idamnjanovic@42 261 % % 'mode', 1, ...
idamnjanovic@42 262 % % 'lambda',Edata*Edata,...
idamnjanovic@42 263 % % 'L', maxatoms);
idamnjanovic@42 264 % % % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 265 % % % but backward compatible with KSVD definition of denoising
idamnjanovic@42 266 % %
idamnjanovic@42 267 % % SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@42 268 % SMALL.solver(3).toolbox='SMALL';
idamnjanovic@42 269 % SMALL.solver(3).name='SMALL_cgp';
idamnjanovic@42 270 % SMALL.solver(3).param=sprintf('%d, %.2f', maxatoms, sqrt(Edata));
idamnjanovic@42 271 % % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 272 % % but backward compatible with KSVD definition of denoising
idamnjanovic@42 273 %
idamnjanovic@42 274 % SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@42 275
idamnjanovic@42 276 % %%
idamnjanovic@42 277 % % Use RLS-DLA
idamnjanovic@42 278 %
idamnjanovic@42 279 % % Initialising Dictionary structure
idamnjanovic@42 280 % % Setting Dictionary structure fields (toolbox, name, param, D and time)
idamnjanovic@42 281 % % to zero values
idamnjanovic@42 282 %
idamnjanovic@42 283 % SMALL.DL(3)=SMALL_init_DL();
idamnjanovic@42 284 %
idamnjanovic@42 285 % % Defining fields needed for dictionary learning
idamnjanovic@42 286 %
idamnjanovic@42 287 % SMALL.DL(3).toolbox = 'mpv2';
idamnjanovic@42 288 % SMALL.DL(3).name = 'rlsdla';
idamnjanovic@42 289 %
idamnjanovic@42 290 % % Type 'help mexTrainDL in MATLAB prompt for explanation of parameters.
idamnjanovic@42 291 %
idamnjanovic@42 292 % SMALL.DL(3).param=struct(...
idamnjanovic@42 293 % 'D', SMALL.Problem.initdict,...
idamnjanovic@42 294 % 'K', SMALL.Problem.p,...
idamnjanovic@42 295 % 'abs', Edata*Edata,...
idamnjanovic@42 296 % 'lambda', 0.995,...
idamnjanovic@42 297 % 'iternum',1);
idamnjanovic@42 298 %
idamnjanovic@42 299 % % Learn the dictionary
idamnjanovic@42 300 %
idamnjanovic@42 301 % SMALL.DL(3) = SMALL_learn(SMALL.Problem, SMALL.DL(3));
idamnjanovic@42 302 %
idamnjanovic@42 303 % % Set SMALL.Problem.A dictionary
idamnjanovic@42 304 % % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@42 305 % % only with Problem structure, ie no direct communication between DL and
idamnjanovic@42 306 % % solver structures)
idamnjanovic@42 307 %
idamnjanovic@42 308 %
idamnjanovic@42 309
idamnjanovic@42 310 %%
idamnjanovic@42 311 % Initialising solver structure
idamnjanovic@42 312 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 313 % reconstructed and time) to zero values
idamnjanovic@42 314 %SMALL.DL(3).D(:,225:256)=0;
idamnjanovic@42 315 SMALL.Problem.A = SMALL.DL(3).D;
idamnjanovic@42 316 SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem);
idamnjanovic@42 317 %maxatoms=32;
idamnjanovic@42 318 SMALL.solver(3)=SMALL_init_solver;
idamnjanovic@42 319
idamnjanovic@42 320 % Defining the parameters needed for denoising
idamnjanovic@42 321
idamnjanovic@42 322 % SMALL.solver(3).toolbox='SPAMS';
idamnjanovic@42 323 % SMALL.solver(3).name='mexLasso';
idamnjanovic@42 324 % SMALL.solver(3).param=struct(...
idamnjanovic@42 325 % 'mode', 1, ...
idamnjanovic@42 326 % 'lambda',Edata*Edata,...
idamnjanovic@42 327 % 'L', maxatoms);
idamnjanovic@42 328 % % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 329 % % but backward compatible with KSVD definition of denoising
idamnjanovic@42 330 %
idamnjanovic@42 331 % SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@42 332
idamnjanovic@42 333 % Initialising solver structure
idamnjanovic@42 334 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 335 % reconstructed and time) to zero values
idamnjanovic@42 336
idamnjanovic@42 337 SMALL.solver(3)=SMALL_init_solver;
idamnjanovic@42 338
idamnjanovic@42 339 % Defining the parameters needed for image denoising
idamnjanovic@42 340
idamnjanovic@42 341 SMALL.solver(3).toolbox='ompbox';
idamnjanovic@42 342 SMALL.solver(3).name='omp2';
idamnjanovic@42 343 SMALL.solver(3).param=struct(...
idamnjanovic@42 344 'epsilon',Edata,...
idamnjanovic@42 345 'maxatoms', maxatoms);
idamnjanovic@42 346 % SMALL.solver(3).toolbox='SPAMS';
idamnjanovic@42 347 % SMALL.solver(3).name='mexLasso';
idamnjanovic@42 348 % SMALL.solver(3).param=struct(...
idamnjanovic@42 349 % 'mode', 2, ...
idamnjanovic@42 350 % 'lambda',40,...
idamnjanovic@42 351 % 'L', maxatoms);
idamnjanovic@42 352
idamnjanovic@42 353 % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 354 % but backward compatible with KSVD definition of denoising
idamnjanovic@42 355
idamnjanovic@42 356 SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@42 357 % Plot results and save midi files
idamnjanovic@42 358 SMALL.solver(3).reconstructed.psnr
idamnjanovic@42 359 % show results %
idamnjanovic@42 360
idamnjanovic@42 361 SMALL_ImgDeNoiseResult(SMALL);
idamnjanovic@42 362 end
idamnjanovic@42 363 results(noise_ind,im_num).psnr.ksvd=SMALL.solver(1).reconstructed.psnr;
idamnjanovic@42 364 results(noise_ind,im_num).psnr.odct=SMALL.solver(2).reconstructed.psnr;
idamnjanovic@42 365 results(noise_ind,im_num).psnr.rlsdla=SMALL.solver(3).reconstructed.psnr;
idamnjanovic@42 366 results(noise_ind,im_num).vmrse.ksvd=SMALL.solver(1).reconstructed.vmrse;
idamnjanovic@42 367 results(noise_ind,im_num).vmrse.odct=SMALL.solver(2).reconstructed.vmrse;
idamnjanovic@42 368 results(noise_ind,im_num).vmrse.rlsdla=SMALL.solver(3).reconstructed.vmrse;
idamnjanovic@42 369 results(noise_ind,im_num).ssim.ksvd=SMALL.solver(1).reconstructed.ssim;
idamnjanovic@42 370 results(noise_ind,im_num).ssim.odct=SMALL.solver(2).reconstructed.ssim;
idamnjanovic@42 371 results(noise_ind,im_num).ssim.rlsdla=SMALL.solver(3).reconstructed.ssim;
idamnjanovic@42 372
idamnjanovic@42 373 results(noise_ind,im_num).time.ksvd=SMALL.solver(1).time+SMALL.DL(1).time;
idamnjanovic@42 374 results(noise_ind,im_num).time.rlsdla.time=SMALL.solver(3).time+SMALL.DL(3).time;
idamnjanovic@42 375 %clear SMALL;
idamnjanovic@42 376 end
idamnjanovic@42 377 end
idamnjanovic@42 378 save results.mat results