Mercurial > hg > smallbox

annotate examples/Image Denoising/SMALL_ImgDenoise_DL_test_KSVDvsRLSDLA.m @ 83:4302a91e6033

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
couple of comment lines added
author Maria Jafari <maria.jafari@eecs.qmul.ac.uk>
date Fri, 01 Apr 2011 12:11:16 +0100
parents 62f20b91d870
children 67aae1283973
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.
maria@83 7 % Two 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.
maria@83 11 % - RLS-DLA - Skretting, K.; Engan, K.; , "Recursive Least Squares
maria@83 12 % Dictionary Learning Algorithm," Signal Processing, IEEE Transactions on,
maria@83 13 % vol.58, no.4, pp.2121-2130, April 2010
idamnjanovic@42 14 %
maria@83 15
maria@83 16
maria@83 17 % Centre for Digital Music, Queen Mary, University of London.
maria@83 18 % This file copyright 2011 Ivan Damnjanovic.
idamnjanovic@42 19 %
maria@83 20 % This program is free software; you can redistribute it and/or
maria@83 21 % modify it under the terms of the GNU General Public License as
maria@83 22 % published by the Free Software Foundation; either version 2 of the
maria@83 23 % License, or (at your option) any later version. See the file
maria@83 24 % COPYING included with this distribution for more information.
maria@83 25 %
idamnjanovic@42 26 %%
idamnjanovic@42 27
idamnjanovic@42 28
idamnjanovic@42 29
idamnjanovic@42 30 % If you want to load the image outside of generateImageDenoiseProblem
idamnjanovic@42 31 % function uncomment following lines. This can be useful if you want to
idamnjanovic@42 32 % denoise more then one image for example.
maria@83 33 % Here we are loading test_image.mat that contains structure with 5 images : lena,
maria@83 34 % barbara,boat, house and peppers.
idamnjanovic@42 35 clear;
idamnjanovic@42 36 TMPpath=pwd;
idamnjanovic@42 37 FS=filesep;
idamnjanovic@42 38 [pathstr1, name, ext, versn] = fileparts(which('SMALLboxSetup.m'));
idamnjanovic@42 39 cd([pathstr1,FS,'data',FS,'images']);
idamnjanovic@42 40 load('test_image.mat');
ivan@77 41 cd(TMPpath);
maria@83 42
maria@83 43 % Deffining the noise levels that we want to test
idamnjanovic@42 44
idamnjanovic@42 45 noise_level=[10 20 25 50 100];
maria@83 46
maria@83 47 % Here we loop through different noise levels and images
maria@83 48
maria@83 49 for noise_ind=2:2
maria@83 50 for im_num=1:1
maria@83 51
idamnjanovic@42 52 % Defining Image Denoising Problem as Dictionary Learning
idamnjanovic@42 53 % Problem. As an input we set the number of training patches.
maria@83 54
idamnjanovic@65 55 SMALL.Problem = generateImageDenoiseProblem(test_image(im_num).i, 40000, '',256, noise_level(noise_ind));
ivan@77 56 SMALL.Problem.name=int2str(im_num);
idamnjanovic@42 57
idamnjanovic@42 58 results(noise_ind,im_num).noisy_psnr=SMALL.Problem.noisy_psnr;
idamnjanovic@42 59
idamnjanovic@42 60 %%
idamnjanovic@42 61 % Use KSVD Dictionary Learning Algorithm to Learn overcomplete dictionary
idamnjanovic@42 62
idamnjanovic@42 63 % Initialising Dictionary structure
idamnjanovic@42 64 % Setting Dictionary structure fields (toolbox, name, param, D and time)
idamnjanovic@42 65 % to zero values
idamnjanovic@42 66
idamnjanovic@42 67 SMALL.DL(1)=SMALL_init_DL();
idamnjanovic@42 68
idamnjanovic@42 69 % Defining the parameters needed for dictionary learning
idamnjanovic@42 70
idamnjanovic@42 71 SMALL.DL(1).toolbox = 'KSVD';
idamnjanovic@42 72 SMALL.DL(1).name = 'ksvd';
idamnjanovic@42 73
idamnjanovic@42 74 % Defining the parameters for KSVD
idamnjanovic@42 75 % In this example we are learning 256 atoms in 20 iterations, so that
idamnjanovic@42 76 % every patch in the training set can be represented with target error in
idamnjanovic@42 77 % L2-norm (EData)
idamnjanovic@42 78 % Type help ksvd in MATLAB prompt for more options.
idamnjanovic@42 79
idamnjanovic@42 80 Edata=sqrt(prod(SMALL.Problem.blocksize)) * SMALL.Problem.sigma * SMALL.Problem.gain;
idamnjanovic@42 81 maxatoms = floor(prod(SMALL.Problem.blocksize)/2);
idamnjanovic@42 82 SMALL.DL(1).param=struct(...
idamnjanovic@42 83 'Edata', Edata,...
idamnjanovic@42 84 'initdict', SMALL.Problem.initdict,...
idamnjanovic@42 85 'dictsize', SMALL.Problem.p,...
idamnjanovic@42 86 'exact', 1, ...
idamnjanovic@42 87 'iternum', 20,...
idamnjanovic@42 88 'memusage', 'high');
idamnjanovic@42 89
idamnjanovic@42 90 % Learn the dictionary
idamnjanovic@42 91
idamnjanovic@42 92 SMALL.DL(1) = SMALL_learn(SMALL.Problem, SMALL.DL(1));
idamnjanovic@42 93
idamnjanovic@42 94 % Set SMALL.Problem.A dictionary
idamnjanovic@42 95 % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@42 96 % only with Problem structure, ie no direct communication between DL and
idamnjanovic@42 97 % solver structures)
idamnjanovic@42 98
idamnjanovic@42 99 SMALL.Problem.A = SMALL.DL(1).D;
idamnjanovic@42 100 SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem);
idamnjanovic@42 101
idamnjanovic@42 102 %%
idamnjanovic@42 103 % Initialising solver structure
idamnjanovic@42 104 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 105 % reconstructed and time) to zero values
idamnjanovic@42 106
idamnjanovic@42 107 SMALL.solver(1)=SMALL_init_solver;
idamnjanovic@42 108
idamnjanovic@42 109 % Defining the parameters needed for image denoising
idamnjanovic@42 110
idamnjanovic@42 111 SMALL.solver(1).toolbox='ompbox';
idamnjanovic@42 112 SMALL.solver(1).name='omp2';
idamnjanovic@42 113 SMALL.solver(1).param=struct(...
idamnjanovic@42 114 'epsilon',Edata,...
idamnjanovic@42 115 'maxatoms', maxatoms);
idamnjanovic@42 116
idamnjanovic@42 117 % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 118 % but backward compatible with KSVD definition of denoising
idamnjanovic@42 119
idamnjanovic@42 120 SMALL.solver(1)=SMALL_solve(SMALL.Problem, SMALL.solver(1));
idamnjanovic@42 121 SMALL.solver(1).reconstructed.psnr
idamnjanovic@42 122 %%
idamnjanovic@42 123 % Use KSVDS Dictionary Learning Algorithm to denoise image
idamnjanovic@42 124
idamnjanovic@42 125 % Initialising solver structure
idamnjanovic@42 126 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 127 % reconstructed and time) to zero values
idamnjanovic@42 128 %
idamnjanovic@42 129 % SMALL.DL(2)=SMALL_init_DL();
idamnjanovic@42 130 %
idamnjanovic@42 131 % % Defining the parameters needed for dictionary learning
idamnjanovic@42 132 %
idamnjanovic@42 133 % SMALL.DL(2).toolbox = 'KSVDS';
idamnjanovic@42 134 % SMALL.DL(2).name = 'ksvds';
idamnjanovic@42 135 %
idamnjanovic@42 136 % % Defining the parameters for KSVDS
idamnjanovic@42 137 % % In this example we are learning 256 atoms in 20 iterations, so that
idamnjanovic@42 138 % % every patch in the training set can be represented with target error in
idamnjanovic@42 139 % % L2-norm (EDataS). We also impose "double sparsity" - dictionary itself
idamnjanovic@42 140 % % has to be sparse in the given base dictionary (Tdict - number of
idamnjanovic@42 141 % % nonzero elements per atom).
idamnjanovic@42 142 % % Type help ksvds in MATLAB prompt for more options.
idamnjanovic@42 143 %
idamnjanovic@42 144 %
idamnjanovic@42 145 % SMALL.DL(2).param=struct(...
idamnjanovic@42 146 % 'Edata', Edata, ...
idamnjanovic@42 147 % 'Tdict', 6,...
idamnjanovic@42 148 % 'stepsize', 1,...
idamnjanovic@42 149 % 'dictsize', SMALL.Problem.p,...
idamnjanovic@42 150 % 'iternum', 20,...
idamnjanovic@42 151 % 'memusage', 'high');
idamnjanovic@42 152 % SMALL.DL(2).param.initA = speye(SMALL.Problem.p);
idamnjanovic@42 153 % SMALL.DL(2).param.basedict{1} = odctdict(8,16);
idamnjanovic@42 154 % SMALL.DL(2).param.basedict{2} = odctdict(8,16);
idamnjanovic@42 155 %
idamnjanovic@42 156 % % Learn the dictionary
idamnjanovic@42 157 %
idamnjanovic@42 158 % SMALL.DL(2) = SMALL_learn(SMALL.Problem, SMALL.DL(2));
idamnjanovic@42 159
idamnjanovic@42 160 % Set SMALL.Problem.A dictionary and SMALL.Problem.basedictionary
idamnjanovic@42 161 % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@42 162 % only with Problem structure, ie no direct communication between DL and
idamnjanovic@42 163 % solver structures)
idamnjanovic@42 164
idamnjanovic@42 165 SMALL.Problem.A = SMALL.Problem.initdict;
idamnjanovic@42 166 % SMALL.Problem.basedict{1} = SMALL.DL(2).param.basedict{1};
idamnjanovic@42 167 % SMALL.Problem.basedict{2} = SMALL.DL(2).param.basedict{2};
idamnjanovic@42 168 SMALL.DL(2).D=SMALL.Problem.initdict;
idamnjanovic@42 169 SparseDict=0;
idamnjanovic@42 170 SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem, SparseDict);
idamnjanovic@42 171
idamnjanovic@42 172 %%
idamnjanovic@42 173 % Initialising solver structure
idamnjanovic@42 174 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 175 % reconstructed and time) to zero values
idamnjanovic@42 176
idamnjanovic@42 177 SMALL.solver(2)=SMALL_init_solver;
idamnjanovic@42 178
idamnjanovic@42 179 % Defining the parameters needed for image denoising
idamnjanovic@42 180
idamnjanovic@42 181 SMALL.solver(2).toolbox='ompbox';
idamnjanovic@42 182 SMALL.solver(2).name='omp2';
idamnjanovic@42 183 SMALL.solver(2).param=struct(...
idamnjanovic@42 184 'epsilon',Edata,...
idamnjanovic@42 185 'maxatoms', maxatoms);
idamnjanovic@42 186
idamnjanovic@42 187 % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 188 % but backward compatible with KSVD definition of denoising
idamnjanovic@42 189 % Pay attention that since implicit base dictionary is used, denoising
idamnjanovic@42 190 % can be much faster then using explicit dictionary in KSVD example.
idamnjanovic@42 191
idamnjanovic@42 192 SMALL.solver(2)=SMALL_solve(SMALL.Problem, SMALL.solver(2));
idamnjanovic@42 193 %%
idamnjanovic@42 194
idamnjanovic@42 195 for i =1:1
idamnjanovic@42 196
idamnjanovic@42 197 X=SMALL.Problem.b1;
idamnjanovic@42 198 X_norm=sqrt(sum(X.^2, 1));
idamnjanovic@42 199 [X_norm_sort, p]=sort(X_norm);
idamnjanovic@42 200 p1=p(X_norm_sort>Edata);
maria@83 201 if size(p1,2)>40000
idamnjanovic@42 202 p2 = randperm(size(p1,2));
idamnjanovic@42 203 p2=sort(p2(1:40000));
idamnjanovic@42 204 size(p2,2)
idamnjanovic@42 205 SMALL.Problem.b=X(:,p1(p2));
idamnjanovic@42 206 else
idamnjanovic@42 207 size(p1,2)
idamnjanovic@42 208 SMALL.Problem.b=X(:,p1);
idamnjanovic@42 209
idamnjanovic@42 210 end
idamnjanovic@42 211
idamnjanovic@42 212 lambda=0.9998
idamnjanovic@42 213
idamnjanovic@42 214 % Use Recursive Least Squares
idamnjanovic@42 215 % to Learn overcomplete dictionary
idamnjanovic@42 216
idamnjanovic@42 217 % Initialising Dictionary structure
idamnjanovic@42 218 % Setting Dictionary structure fields (toolbox, name, param, D and time)
idamnjanovic@42 219 % to zero values
idamnjanovic@42 220
idamnjanovic@42 221 SMALL.DL(3)=SMALL_init_DL();
idamnjanovic@42 222
idamnjanovic@42 223 % Defining fields needed for dictionary learning
idamnjanovic@42 224
idamnjanovic@42 225 SMALL.DL(3).toolbox = 'SMALL';
idamnjanovic@42 226 SMALL.DL(3).name = 'SMALL_rlsdla';
idamnjanovic@42 227 SMALL.DL(3).param=struct(...
idamnjanovic@42 228 'Edata', Edata,...
idamnjanovic@42 229 'initdict', SMALL.Problem.initdict,...
idamnjanovic@42 230 'dictsize', SMALL.Problem.p,...
idamnjanovic@42 231 'forgettingMode', 'FIX',...
idamnjanovic@42 232 'forgettingFactor', lambda);
idamnjanovic@42 233
idamnjanovic@42 234 % % Type 'help mexTrainDL in MATLAB prompt for explanation of parameters.
idamnjanovic@42 235 %
idamnjanovic@42 236 % SMALL.DL(3).param=struct(...
idamnjanovic@42 237 % 'D', SMALL.Problem.initdict,...
idamnjanovic@42 238 % 'K', SMALL.Problem.p,...
idamnjanovic@42 239 % 'lambda', 2,...
idamnjanovic@42 240 % 'iter', 200,...
idamnjanovic@42 241 % 'mode', 3, ...
idamnjanovic@42 242 % 'modeD', 0);
idamnjanovic@42 243
idamnjanovic@42 244 % Learn the dictionary
idamnjanovic@42 245
idamnjanovic@42 246 SMALL.DL(3) = SMALL_learn(SMALL.Problem, SMALL.DL(3));
idamnjanovic@42 247 %SMALL.DL(3).D(:,1)=SMALL.DL(1).D(:,1);
idamnjanovic@42 248 %
idamnjanovic@42 249 % % Set SMALL.Problem.A dictionary
idamnjanovic@42 250 % % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@42 251 % % only with Problem structure, ie no direct communication between DL and
idamnjanovic@42 252 % % solver structures)
idamnjanovic@42 253 %
idamnjanovic@42 254 %
idamnjanovic@42 255 %
idamnjanovic@42 256 % %%
idamnjanovic@42 257 % % Initialising solver structure
idamnjanovic@42 258 % % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 259 % % reconstructed and time) to zero values
idamnjanovic@42 260 % SMALL.Problem.A = SMALL.DL(1).D;
idamnjanovic@42 261 % SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem);
idamnjanovic@42 262 % maxatoms=5;
idamnjanovic@42 263 % SMALL.solver(3)=SMALL_init_solver;
idamnjanovic@42 264 %
idamnjanovic@42 265 % % Defining the parameters needed for denoising
idamnjanovic@42 266 %
idamnjanovic@42 267 % % SMALL.solver(3).toolbox='SPAMS';
idamnjanovic@42 268 % % SMALL.solver(3).name='mexLasso';
idamnjanovic@42 269 % % SMALL.solver(3).param=struct(...
idamnjanovic@42 270 % % 'mode', 1, ...
idamnjanovic@42 271 % % 'lambda',Edata*Edata,...
idamnjanovic@42 272 % % 'L', maxatoms);
idamnjanovic@42 273 % % % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 274 % % % but backward compatible with KSVD definition of denoising
idamnjanovic@42 275 % %
idamnjanovic@42 276 % % SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@42 277 % SMALL.solver(3).toolbox='SMALL';
idamnjanovic@42 278 % SMALL.solver(3).name='SMALL_cgp';
idamnjanovic@42 279 % SMALL.solver(3).param=sprintf('%d, %.2f', maxatoms, sqrt(Edata));
idamnjanovic@42 280 % % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 281 % % but backward compatible with KSVD definition of denoising
idamnjanovic@42 282 %
idamnjanovic@42 283 % SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@42 284
idamnjanovic@42 285 % %%
idamnjanovic@42 286 % % Use RLS-DLA
idamnjanovic@42 287 %
idamnjanovic@42 288 % % Initialising Dictionary structure
idamnjanovic@42 289 % % Setting Dictionary structure fields (toolbox, name, param, D and time)
idamnjanovic@42 290 % % to zero values
idamnjanovic@42 291 %
idamnjanovic@42 292 % SMALL.DL(3)=SMALL_init_DL();
idamnjanovic@42 293 %
idamnjanovic@42 294 % % Defining fields needed for dictionary learning
idamnjanovic@42 295 %
idamnjanovic@42 296 % SMALL.DL(3).toolbox = 'mpv2';
idamnjanovic@42 297 % SMALL.DL(3).name = 'rlsdla';
idamnjanovic@42 298 %
idamnjanovic@42 299 % % Type 'help mexTrainDL in MATLAB prompt for explanation of parameters.
idamnjanovic@42 300 %
idamnjanovic@42 301 % SMALL.DL(3).param=struct(...
idamnjanovic@42 302 % 'D', SMALL.Problem.initdict,...
idamnjanovic@42 303 % 'K', SMALL.Problem.p,...
idamnjanovic@42 304 % 'abs', Edata*Edata,...
idamnjanovic@42 305 % 'lambda', 0.995,...
idamnjanovic@42 306 % 'iternum',1);
idamnjanovic@42 307 %
idamnjanovic@42 308 % % Learn the dictionary
idamnjanovic@42 309 %
idamnjanovic@42 310 % SMALL.DL(3) = SMALL_learn(SMALL.Problem, SMALL.DL(3));
idamnjanovic@42 311 %
idamnjanovic@42 312 % % Set SMALL.Problem.A dictionary
idamnjanovic@42 313 % % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@42 314 % % only with Problem structure, ie no direct communication between DL and
idamnjanovic@42 315 % % solver structures)
idamnjanovic@42 316 %
idamnjanovic@42 317 %
idamnjanovic@42 318
idamnjanovic@42 319 %%
idamnjanovic@42 320 % Initialising solver structure
idamnjanovic@42 321 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 322 % reconstructed and time) to zero values
idamnjanovic@42 323 %SMALL.DL(3).D(:,225:256)=0;
idamnjanovic@42 324 SMALL.Problem.A = SMALL.DL(3).D;
idamnjanovic@42 325 SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem);
idamnjanovic@42 326 %maxatoms=32;
idamnjanovic@42 327 SMALL.solver(3)=SMALL_init_solver;
idamnjanovic@42 328
idamnjanovic@42 329 % Defining the parameters needed for denoising
idamnjanovic@42 330
idamnjanovic@42 331 % SMALL.solver(3).toolbox='SPAMS';
idamnjanovic@42 332 % SMALL.solver(3).name='mexLasso';
idamnjanovic@42 333 % SMALL.solver(3).param=struct(...
idamnjanovic@42 334 % 'mode', 1, ...
idamnjanovic@42 335 % 'lambda',Edata*Edata,...
idamnjanovic@42 336 % 'L', maxatoms);
idamnjanovic@42 337 % % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 338 % % but backward compatible with KSVD definition of denoising
idamnjanovic@42 339 %
idamnjanovic@42 340 % SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@42 341
idamnjanovic@42 342 % Initialising solver structure
idamnjanovic@42 343 % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@42 344 % reconstructed and time) to zero values
idamnjanovic@42 345
idamnjanovic@42 346 SMALL.solver(3)=SMALL_init_solver;
idamnjanovic@42 347
idamnjanovic@42 348 % Defining the parameters needed for image denoising
idamnjanovic@42 349
idamnjanovic@42 350 SMALL.solver(3).toolbox='ompbox';
idamnjanovic@42 351 SMALL.solver(3).name='omp2';
idamnjanovic@42 352 SMALL.solver(3).param=struct(...
idamnjanovic@42 353 'epsilon',Edata,...
idamnjanovic@42 354 'maxatoms', maxatoms);
idamnjanovic@42 355 % SMALL.solver(3).toolbox='SPAMS';
idamnjanovic@42 356 % SMALL.solver(3).name='mexLasso';
idamnjanovic@42 357 % SMALL.solver(3).param=struct(...
idamnjanovic@42 358 % 'mode', 2, ...
idamnjanovic@42 359 % 'lambda',40,...
idamnjanovic@42 360 % 'L', maxatoms);
idamnjanovic@42 361
idamnjanovic@42 362 % Denoising the image - SMALL_denoise function is similar to SMALL_solve,
idamnjanovic@42 363 % but backward compatible with KSVD definition of denoising
idamnjanovic@42 364
idamnjanovic@42 365 SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@42 366 % Plot results and save midi files
idamnjanovic@42 367 SMALL.solver(3).reconstructed.psnr
idamnjanovic@42 368 % show results %
idamnjanovic@42 369
idamnjanovic@42 370 SMALL_ImgDeNoiseResult(SMALL);
idamnjanovic@42 371 end
idamnjanovic@42 372 results(noise_ind,im_num).psnr.ksvd=SMALL.solver(1).reconstructed.psnr;
idamnjanovic@42 373 results(noise_ind,im_num).psnr.odct=SMALL.solver(2).reconstructed.psnr;
idamnjanovic@42 374 results(noise_ind,im_num).psnr.rlsdla=SMALL.solver(3).reconstructed.psnr;
idamnjanovic@42 375 results(noise_ind,im_num).vmrse.ksvd=SMALL.solver(1).reconstructed.vmrse;
idamnjanovic@42 376 results(noise_ind,im_num).vmrse.odct=SMALL.solver(2).reconstructed.vmrse;
idamnjanovic@42 377 results(noise_ind,im_num).vmrse.rlsdla=SMALL.solver(3).reconstructed.vmrse;
idamnjanovic@42 378 results(noise_ind,im_num).ssim.ksvd=SMALL.solver(1).reconstructed.ssim;
idamnjanovic@42 379 results(noise_ind,im_num).ssim.odct=SMALL.solver(2).reconstructed.ssim;
idamnjanovic@42 380 results(noise_ind,im_num).ssim.rlsdla=SMALL.solver(3).reconstructed.ssim;
idamnjanovic@42 381
idamnjanovic@42 382 results(noise_ind,im_num).time.ksvd=SMALL.solver(1).time+SMALL.DL(1).time;
idamnjanovic@42 383 results(noise_ind,im_num).time.rlsdla.time=SMALL.solver(3).time+SMALL.DL(3).time;
idamnjanovic@42 384 %clear SMALL;
idamnjanovic@42 385 end
idamnjanovic@42 386 end
idamnjanovic@42 387 save results.mat results