]> AND Private Git Repository - these_gilles.git/blob - BM3D/VBM3D.m
Logo AND Algorithmique Numérique Distribuée

Private GIT Repository
20 déc
[these_gilles.git] / BM3D / VBM3D.m
1 function [PSNR_FINAL_ESTIMATE, y_hat_wi] = VBM3D(Xnoisy, sigma, NumberOfFrames, dump_information, Xorig, bm3dProfile)
2 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
3 %
4 %  VBM3D is a Matlab function for attenuation of additive white Gaussian 
5 %  noise from grayscale videos. This algorithm reproduces the results from the article:
6 %
7 %  [1] K. Dabov, A. Foi, and K. Egiazarian, "Video denoising by sparse 3D
8 %  transform-domain collaborative filtering," European Signal Processing
9 %  Conference (EUSIPCO-2007), September 2007. (accepted)
10 %
11 %  INTERFACE:
12 %
13 %  [PSNR, Xest] = VBM3D(Xnoisy, Sigma, NFrames, PrintInfo, Xorig)
14 %
15 %  INPUTS:
16 %   1)  Xnoisy     --> A filename of a noisy .avi video, e.g. Xnoisy = 'gstennisg20.avi'
17 %        OR
18 %       Xnoisy     --> A 3D matrix of a noisy video in a  (floating point data in range [0,1],
19 %                                                     or in [0,255])
20 %   2)  Sigma --> Noise standard deviation (assumed range is [0,255], no matter what is
21 %                                           the input's range)
22 %
23 %   3)  NFrames   (optional paremter!) --> Number of frames to process. If set to 0 or 
24 %                                          ommited, then process all frames (default: 0).
25 %
26 %   4)  PrintInfo (optional paremter!) --> If non-zero, then print to screen and save 
27 %                                          the denoised video in .AVI
28 %                                          format. (default: 1)
29 %
30 %   5)  Xorig     (optional paremter!) --> Original video's filename or 3D matrix 
31 %                                          If provided, PSNR, ISNR will be computed.
32 %
33 %   NOTE: If Xorig == Xnoisy, then artificial noise is added internally and the
34 %   obtained noisy video is denoised.
35 %
36 %  OUTPUTS:
37 %  
38 %   1) PSNR --> If Xorig is valid video, then this contains the PSNR of the
39 %                denoised one
40 %
41 %   1) Xest --> Final video estimate in a 3D matrix (intensities in range [0,1])
42 %
43 %   *) If "PrintInfo" is non-zero, then save the denoised video in the current 
44 %       MATLAB folder.
45 %
46 %  USAGE EXAMPLES:
47 %
48 %     1) Denoise a noisy (clipped in [0,255] range) video sequence, e.g. 
49 %        'gsalesmang20.avi' corrupted with AWGN with std. dev. 20:
50 %          
51 %          Xest = VBM3D('gsalesmang20.avi', 20, 0, 1); 
52 %     
53 %     2) The same, but also print PSNR, ISNR numbers.
54 %        
55 %          Xest = VBM3D('gsalesmang20.avi', 20, 0, 1, 'gsalesman.avi');
56 %
57 %     3) Add artificial noise to a video, then denoise it (without 
58 %        considering clipping in [0,255]):
59 %        
60 %          Xest = VBM3D('gsalesman.avi', 20, 0, 1, 'gsalesman.avi');
61 %  
62 %
63 %  RESTRICTIONS:
64 %
65 %     Since the video sequences are read into memory as 3D matrices,
66 %     there apply restrictions on the input video size, which are thus
67 %     proportional to the maximum memory allocatable by Matlab.
68 %
69 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
70 %
71 % Copyright © 2007 Tampere University of Technology. All rights reserved.
72 % This work should only be used for nonprofit purposes.
73 %
74 % AUTHORS:
75 %     Kostadin Dabov, email: dabov _at_ cs.tut.fi
76 %
77 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
78
79 % If no input argument is provided, then use these internal ones:
80 if exist('sigma', 'var') ~= 1,
81     Xnoisy = 'gsalesmang20.avi'; Xorig = 'gsalesman.avi'; sigma = 20;
82     %Xnoisy = 'gstennisg20.avi';  Xorig = 'gstennis.avi';  sigma = 20;
83     %Xnoisy = 'gflowersg20.avi';   Xorig = 'gflower.avi';   sigma = 20;
84     
85     %Xnoisy = 'gsalesman.avi'; Xorig = Xnoisy; sigma = 20;
86     
87     NumberOfFrames = 0; %% 0 means process ALL frames.
88 end
89
90
91
92 if exist('dump_information', 'var') ~= 1,
93     dump_information = 1; % 1 -> print informaion to the screen and save the processed video as an AVI file
94 end
95
96 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
97 %%%% Obtain infromation about the input noisy video
98 %%%%
99 if (ischar(Xnoisy) == 1), % if the input is a video filename
100     isCharacterName = 1;
101     Xnoisy_name = Xnoisy;
102     videoInfo = aviinfo(Xnoisy);
103     videoHeight = videoInfo.Height;
104     videoWidth = videoInfo.Width;
105     TotalFrames = videoInfo.NumFrames;
106 elseif length(size(Xnoisy)) == 3% the input argument is a 3D video (spatio-temporal) matrix
107     Xnoisy_name = 'Input 3D matrix';
108     isCharacterName = 0;
109     [videoHeight, videoWidth, TotalFrames] = size(Xnoisy);
110 else
111     fprintf('Oops! The input argument Xnoisy should be either a filename or a 3D matrix!\n');
112     PSNR_FINAL_ESTIMATE = 0;
113     y_hat_wi = 0;
114     return;
115 end
116
117 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
118 %%%% Check if we want to process all frames, and save as 'NumberOfFrames' 
119 %%%% the desired number of frames to process
120 %%%%
121 if exist('NumberOfFrames', 'var') == 1,
122     if NumberOfFrames <= 0,
123         NumberOfFrames = TotalFrames;
124     else
125         NumberOfFrames = max(min(NumberOfFrames, TotalFrames), 1);
126     end    
127 else
128     NumberOfFrames = TotalFrames;
129 end
130
131 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
132 %%%%  Quality/complexity trade-off
133 %%%%
134 %%%%  'np' --> Normal Profile (balanced quality)
135 %%%%  'lc' --> Low Complexity Profile (fast, lower quality)
136 %%%%
137 if (exist('bm3dProfile', 'var') ~= 1)
138     bm3dProfile         = 'np';
139 end
140
141 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
142 %%%% Parameters for the Normal Profile.
143 %%%%
144 %%%% Select transforms ('dct', 'dst', 'hadamard', or anything that is listed by 'help wfilters'):
145 transform_2D_HT_name     = 'bior1.5'; %% transform used for the HT filt. of size N1 x N1
146 transform_2D_Wiener_name = 'dct';     %% transform used for the Wiener filt. of size N1_wiener x N1_wiener
147 transform_3rd_dim_name   = 'haar'; %% tranform used in the 3-rd dim, the same for HT and Wiener filt.
148
149 %%%% Step 1: Hard-thresholding (HT) parameters:
150 denoiseFrames       = min(9, NumberOfFrames); % number of frames in the temporalwindow (should not exceed the total number of frames 'NumberOfFrames')
151 N1                  = 8;  %% N1 x N1 is the block size used for the hard-thresholding (HT) filtering
152 Nstep               = 6;  %% sliding step to process every next refernece block
153 N2                  = 8;  %% maximum number of similar blocks (maximum size of the 3rd dimension of the 3D groups)
154 Ns                  = 7;  %% length of the side of the search neighborhood for full-search block-matching (BM)
155 Npr                 = 5;  %% length of the side of the motion-adaptive search neighborhood, use din the predictive-search BM
156 tau_match           = 3000; %% threshold for the block distance (d-distance)
157 lambda_thr3D        = 2.7; %% threshold parameter for the hard-thresholding in 3D DFT domain
158 dsub                = 7;  %% a small value subtracted from the distnce of blocks with the same spatial coordinate as the reference one 
159 Nb                  = 2;  %% number of blocks to follow in each next frame, used in the predictive-search BM
160 beta                = 2.0; %% the beta parameter of the 2D Kaiser window used in the reconstruction
161
162 %%%% Step 2: Wiener filtering parameters:
163 denoiseFramesW      = min(9, NumberOfFrames);
164 N1_wiener           = 7;
165 Nstep_wiener        = 4;
166 N2_wiener           = 8;
167 Ns_wiener           = 7;
168 Npr_wiener          = 5;
169 tau_match_wiener    = 1500;
170 beta_wiener         = 2.0;
171 dsub_wiener         = 3;
172 Nb_wiener           = 2;
173
174 %%%% Block-matching parameters:
175 stepFS              = 1; %% step that forces to switch to full-search BM, "1" implies always full-search
176 smallLN             = 3; %% if stepFS > 1, then this specifies the size of the small local search neighb.
177 stepFSW             = 1;
178 smallLNW            = 3;
179 thrToIncStep        = 8;  %% used in the HT filtering to increase the sliding step in uniform regions
180
181
182 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
183 %%%% Parameters for the Low Complexity Profile.
184 %%%%
185 if strcmp(bm3dProfile, 'lc') == 1,
186     lambda_thr3D = 2.8;
187     smallLN   = 2;
188     smallLNW  = 2;
189     denoiseFrames  = min(5, NumberOfFrames);
190     denoiseFramesW = min(5, NumberOfFrames);
191     N2_wiener = 4;
192     N2 = 4;
193     Ns = 3;
194     Ns_wiener = 3;
195     NB = 1;
196     Nb_wiener = 1;
197 end
198
199 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
200 %%%% Parameters for the High Profile.
201 %%%%
202 if strcmp(bm3dProfile, 'hi') == 1,
203     Nstep        = 3;
204     Nstep_wiener = 3;
205 end
206
207 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
208 %%%% Parameters for the "Very Noisy" Profile.
209 %%%%
210 if sigma > 30,
211     N1 = 8;
212     N1_wiener = 8;
213     Nstep = 6;
214     tau_match    = 4500;
215     tau_match_wiener    = 3000;
216 end
217
218
219 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
220 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
221 %%%% Note: touch below this point only if you know what you are doing!
222 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
223 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
224
225 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
226 %%%% Extract the input noisy video and make sure intensities are in [0,1]
227 %%%% interval, using single-precision float
228 if isCharacterName,
229     mno = aviread(Xnoisy_name);
230     z = zeros([videoHeight, videoWidth, NumberOfFrames], 'single');
231     for cf = 1:NumberOfFrames
232         z(:,:,cf) = single(mno(cf).cdata(:,:,1)) * 0.0039216; % 1/255 = 0.0039216
233     end
234     clear  mno
235 else
236     if isinteger(Xnoisy) == 1,
237         z = single(Xnoisy) * 0.0039216; % 1/255 = 0.0039216
238     elseif isfloat(Xnoisy) == 0,
239         fprintf('Unknown format of "Xnoisy"! Must be a filename (array of char) or a 3D array of either floating point data (range [0,1]) or integer data (range [0,255]). \n');
240         return;
241     else        
242         z = single(Xnoisy);
243     end
244 end
245
246 clear Xnoisy;
247
248 %%%% If the original video is provided, then extract it to 'Xorig' 
249 %%%% which is later used to compute PSNR and ISNR
250 if exist('Xorig', 'var') == 1,
251     randn('seed', 0);
252     if ischar(Xorig) == 0,
253         if isinteger(Xorig) == 1,
254             y = single(Xorig) * 0.0039216; % 1/255 = 0.0039216
255         elseif isfloat(Xorig) == 0,
256             fprintf('Unknown format of "Xorig"! Must be a filename (array of char) or a 3D array of either floating point data (range [0,1]) or integer data (range [0,255]). \n');
257             return;            
258         else
259             y = single(Xorig);
260         end
261     else        
262         if strcmp(Xorig, Xnoisy_name) == 1, %% special case, noise is aritifically added
263             y = z;
264             z = z + (sigma/255) * randn(size(z));
265         else
266             mo = aviread(Xorig);
267             y = zeros([videoHeight, videoWidth, NumberOfFrames], 'single');
268             for cf = 1:NumberOfFrames
269                 y(:,:,cf) = single(mo(cf).cdata(:,:,1)) * 0.0039216; % 1/255 = 0.0039216
270             end
271             clear mo
272         end
273     end
274 end
275
276 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
277 %%%% Create transform matrices, etc.
278 %%%%
279 decLevel           = 0;    %% dec. levels of the dyadic wavelet 2D transform for blocks (0 means full decomposition, higher values decrease the dec. number)
280 decLevel3          = 0;    %% dec. level for the wavelet transform in the 3rd dimension
281
282 [Tfor, Tinv]       = getTransfMatrix(N1, transform_2D_HT_name, decLevel); %% get (normalized) forward and inverse transform matrices
283 [TforW, TinvW]     = getTransfMatrix(N1_wiener, transform_2D_Wiener_name); %% get (normalized) forward and inverse transform matrices
284 thr_mask           = ones(N1); %% N1xN1 mask of threshold scaling coeff. --- by default there is no scaling, however the use of different thresholds for different wavelet decompoistion subbands can be done with this matrix
285
286 if (strcmp(transform_3rd_dim_name, 'haar') == 1 || strcmp(transform_3rd_dim_name(end-2:end), '1.1') == 1),
287     %%% Fast internal transform is used, no need to generate transform
288     %%% matrices.
289     hadper_trans_single_den         = {};
290     inverse_hadper_trans_single_den = {};
291 else
292     %%% Create transform matrices. The transforms are later computed by
293     %%% matrix multiplication with them
294     for hh = [1 2 4 8 16 32];
295         [Tfor3rd, Tinv3rd]   = getTransfMatrix(hh, transform_3rd_dim_name, decLevel3);
296         hadper_trans_single_den{hh}         = single(Tfor3rd);
297         inverse_hadper_trans_single_den{hh} = single(Tinv3rd');
298     end
299 end
300
301 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
302 %%%% 2D Kaiser windows that scale the reconstructed blocks
303 %%%%
304 if beta_wiener==2 & beta==2 & N1_wiener==7 & N1==8 % hardcode the window function so that the signal processing toolbox is not needed by default
305     Wwin2D = [ 0.1924    0.2989    0.3846    0.4325    0.4325    0.3846    0.2989    0.1924;
306         0.2989    0.4642    0.5974    0.6717    0.6717    0.5974    0.4642    0.2989;
307         0.3846    0.5974    0.7688    0.8644    0.8644    0.7688    0.5974    0.3846;
308         0.4325    0.6717    0.8644    0.9718    0.9718    0.8644    0.6717    0.4325;
309         0.4325    0.6717    0.8644    0.9718    0.9718    0.8644    0.6717    0.4325;
310         0.3846    0.5974    0.7688    0.8644    0.8644    0.7688    0.5974    0.3846;
311         0.2989    0.4642    0.5974    0.6717    0.6717    0.5974    0.4642    0.2989;
312         0.1924    0.2989    0.3846    0.4325    0.4325    0.3846    0.2989    0.1924 ];
313     Wwin2D_wiener = [ 0.1924    0.3151    0.4055    0.4387    0.4055    0.3151    0.1924;
314         0.3151    0.5161    0.6640    0.7184    0.6640    0.5161    0.3151;
315         0.4055    0.6640    0.8544    0.9243    0.8544    0.6640    0.4055;
316         0.4387    0.7184    0.9243    1.0000    0.9243    0.7184    0.4387;
317         0.4055    0.6640    0.8544    0.9243    0.8544    0.6640    0.4055;
318         0.3151    0.5161    0.6640    0.7184    0.6640    0.5161    0.3151;
319         0.1924    0.3151    0.4055    0.4387    0.4055    0.3151    0.1924 ];
320 else
321     Wwin2D           = kaiser(N1, beta) * kaiser(N1, beta)'; % Kaiser window used in the aggregation of the HT part
322     Wwin2D_wiener    = kaiser(N1_wiener, beta_wiener) * kaiser(N1_wiener, beta_wiener)'; % Kaiser window used in the aggregation of the Wiener filt. part
323 end
324
325 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
326 %%%% Read an image, generate noise and add it to the image
327 %%%%
328
329 l2normLumChrom = ones(NumberOfFrames,1); %%% NumberOfFrames == nSl !
330
331 if dump_information == 1,
332     fprintf('Video: %s (%dx%dx%d), sigma: %.1f\n', Xnoisy_name, videoHeight, videoWidth, NumberOfFrames, sigma);
333 end
334
335 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
336 %%%% Initial estimate by hard-thresholding filtering
337 tic;
338 y_hat = bm3d_thr_video(z, hadper_trans_single_den, Nstep, N1, N2, 0,...
339     lambda_thr3D, tau_match*N1*N1/(255*255), (Ns-1)/2, sigma/255, thrToIncStep, single(Tfor), single(Tinv)', inverse_hadper_trans_single_den, single(thr_mask), 'unused arg', dsub*dsub/255, l2normLumChrom, Wwin2D, (Npr-1)/2, stepFS, denoiseFrames, Nb );
340 estimate_elapsed_time = toc;
341
342 if exist('Xorig', 'var') == 1,
343     PSNR_INITIAL_ESTIMATE = 10*log10(1/mean((double(y(:))-double(y_hat(:))).^2));
344     PSNR_NOISE = 10*log10(1/mean((double(y(:))-double(z(:))).^2));
345     ISNR_INITIAL_ESTIMATE = PSNR_INITIAL_ESTIMATE - PSNR_NOISE;
346
347     if dump_information == 1,    
348         fprintf('BASIC ESTIMATE (time: %.1f sec), PSNR: %.3f dB, ISNR: %.3f dB\n', ...
349             estimate_elapsed_time, PSNR_INITIAL_ESTIMATE, ISNR_INITIAL_ESTIMATE);
350     end
351 end
352       
353
354 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
355 % %%%% Final estimate by Wiener filtering (using the hard-thresholding
356 % initial estimate)
357 tic;
358 y_hat_wi = bm3d_wiener_video(z, y_hat, hadper_trans_single_den, Nstep_wiener, N1_wiener, N2_wiener, ...
359     'unused_arg', tau_match_wiener*N1_wiener*N1_wiener/(255*255), (Ns_wiener-1)/2, sigma/255, 'unused arg', single(TforW), single(TinvW)', inverse_hadper_trans_single_den, 'unused arg', dsub_wiener*dsub_wiener/255, l2normLumChrom, Wwin2D_wiener, (Npr_wiener-1)/2, stepFSW, denoiseFramesW, Nb_wiener );
360
361 % In case the input noisy video is clipped in [0,1], then apply declipping  
362 if isCharacterName
363     if exist('Xorig', 'var') == 1
364         if ~strcmp(Xorig, Xnoisy_name)
365             [y_hat_wi] = ClipComp16b(sigma/255, y_hat_wi);
366         end
367     else
368         [y_hat_wi] = ClipComp16b(sigma/255, y_hat_wi);
369     end
370 end
371
372 wiener_elapsed_time = toc;
373
374
375
376 PSNR_FINAL_ESTIMATE = 0;
377 if exist('Xorig', 'var') == 1,
378     PSNR_FINAL_ESTIMATE = 10*log10(1/mean((double(y(:))-double(y_hat_wi(:))).^2)); 
379     ISNR_FINAL_ESTIMATE = PSNR_FINAL_ESTIMATE - 10*log10(1/mean((double(y(:))-double(z(:))).^2));
380 end
381
382 if dump_information == 1,
383
384     text_psnr = '';
385     if exist('Xorig', 'var') == 1
386
387         
388         %%%% Un-comment the following to print the PSNR of each frame
389         %
390         %     PSNRs = zeros(NumberOfFrames,1);
391         %     for ii = [1:NumberOfFrames],
392         %         PSNRs(ii) = 10*log10(1/mean2((y(:,:,ii)-y_hat_wi(:,:,ii)).^2));
393         %         fprintf(['Frame: ' sprintf('%d',ii) ', PSNR: ' sprintf('%.2f',PSNRs(ii)) '\n']);
394         %     end
395         %
396
397         fprintf('FINAL ESTIMATE, PSNR: %.3f dB, ISNR: %.3f dB\n', ...
398              PSNR_FINAL_ESTIMATE, ISNR_FINAL_ESTIMATE);
399
400         figure, imshow(double(z(:,:,ceil(NumberOfFrames/2)))); % show the central frame
401         title(sprintf('Noisy frame #%d',ceil(NumberOfFrames/2)));           
402         
403         figure, imshow(double(y_hat_wi(:,:,ceil(NumberOfFrames/2)))); % show the central frame
404         title(sprintf('Denoised frame #%d',ceil(NumberOfFrames/2)));
405         
406         text_psnr = sprintf('_PSNR%.2f', PSNR_FINAL_ESTIMATE);
407     end
408     
409     fprintf('The denoising took: %.1f sec (%.4f sec/frame). ', ...
410         wiener_elapsed_time+estimate_elapsed_time, (wiener_elapsed_time+estimate_elapsed_time)/NumberOfFrames);
411
412     
413     text_vid = 'Denoised';
414     FRATE = 30; % default value
415     if isCharacterName,
416         text_vid = Xnoisy_name(1:end-4);
417         ainfo = aviinfo(Xnoisy_name);
418         FRATE = ainfo.FramesPerSecond;
419     end
420
421     avi_filename = sprintf('%s%s_%s_BM3D.avi', text_vid, text_psnr, bm3dProfile);
422     
423     if exist(avi_filename, 'file') ~= 0,
424         delete(avi_filename);
425     end
426     mov = avifile(avi_filename, 'Colormap', gray(256), 'compression', 'None', 'fps', FRATE);
427     for ii = [1:NumberOfFrames],
428         mov = addframe(mov, uint8(round(255*double(y_hat_wi(:,:,ii)))));
429     end
430     mov = close(mov);
431     fprintf('The denoised video written to: %s.\n\n', avi_filename);
432     
433 end
434
435 return;
436
437
438
439
440
441 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
442 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
443 % Some auxiliary functions 
444 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
445 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
446
447
448
449
450 function [Tforward, Tinverse] = getTransfMatrix (N, transform_type, dec_levels)
451 %
452 % Create forward and inverse transform matrices, which allow for perfect
453 % reconstruction. The forward transform matrix is normalized so that the 
454 % l2-norm of each basis element is 1.
455 %
456 % [Tforward, Tinverse] = getTransfMatrix (N, transform_type, dec_levels)
457 %
458 %  INPUTS:
459 %
460 %   N               --> Size of the transform (for wavelets, must be 2^K)
461 %
462 %   transform_type  --> 'dct', 'dst', 'hadamard', or anything that is 
463 %                       listed by 'help wfilters' (bi-orthogonal wavelets)
464 %                       'DCrand' -- an orthonormal transform with a DC and all
465 %                       the other basis elements of random nature
466 %
467 %   dec_levels      --> If a wavelet transform is generated, this is the
468 %                       desired decomposition level. Must be in the
469 %                       range [0, log2(N)-1], where "0" implies
470 %                       full decomposition.
471 %
472 %  OUTPUTS:
473 %
474 %   Tforward        --> (N x N) Forward transform matrix
475 %
476 %   Tinverse        --> (N x N) Inverse transform matrix
477 %
478
479 if exist('dec_levels', 'var') ~= 1,
480     dec_levels = 0;
481 end
482
483 if N == 1,
484     Tforward = 1;
485 elseif strcmp(transform_type, 'hadamard') == 1,
486     Tforward    = hadamard(N);
487 elseif (N == 8) & strcmp(transform_type, 'bior1.5')==1 % hardcoded transform so that the wavelet toolbox is not needed to generate it
488     Tforward =  [ 0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274;
489        0.219417649252501   0.449283757993216   0.449283757993216   0.219417649252501  -0.219417649252501  -0.449283757993216  -0.449283757993216  -0.219417649252501;
490        0.569359398342846   0.402347308162278  -0.402347308162278  -0.569359398342846  -0.083506045090284   0.083506045090284  -0.083506045090284   0.083506045090284;
491       -0.083506045090284   0.083506045090284  -0.083506045090284   0.083506045090284   0.569359398342846   0.402347308162278  -0.402347308162278  -0.569359398342846;
492        0.707106781186547  -0.707106781186547                   0                   0                   0                   0                   0                   0;
493                        0                   0   0.707106781186547  -0.707106781186547                   0                   0                   0                   0;
494                        0                   0                   0                   0   0.707106781186547  -0.707106781186547                   0                   0;
495                        0                   0                   0                   0                   0                   0   0.707106781186547  -0.707106781186547];   
496 elseif (N == 8) & strcmp(transform_type, 'dct')==1 % hardcoded transform so that the signal processing toolbox is not needed to generate it
497     Tforward = [ 0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274   0.353553390593274;
498        0.490392640201615   0.415734806151273   0.277785116509801   0.097545161008064  -0.097545161008064  -0.277785116509801  -0.415734806151273  -0.490392640201615;
499        0.461939766255643   0.191341716182545  -0.191341716182545  -0.461939766255643  -0.461939766255643  -0.191341716182545   0.191341716182545   0.461939766255643;
500        0.415734806151273  -0.097545161008064  -0.490392640201615  -0.277785116509801   0.277785116509801   0.490392640201615   0.097545161008064  -0.415734806151273;
501        0.353553390593274  -0.353553390593274  -0.353553390593274   0.353553390593274   0.353553390593274  -0.353553390593274  -0.353553390593274   0.353553390593274;
502        0.277785116509801  -0.490392640201615   0.097545161008064   0.415734806151273  -0.415734806151273  -0.097545161008064   0.490392640201615  -0.277785116509801;
503        0.191341716182545  -0.461939766255643   0.461939766255643  -0.191341716182545  -0.191341716182545   0.461939766255643  -0.461939766255643   0.191341716182545;
504        0.097545161008064  -0.277785116509801   0.415734806151273  -0.490392640201615   0.490392640201615  -0.415734806151273   0.277785116509801  -0.097545161008064];
505 elseif (N == 8) & strcmp(transform_type, 'dst')==1 % hardcoded transform so that the PDE toolbox is not needed to generate it
506     Tforward = [ 0.161229841765317   0.303012985114696   0.408248290463863   0.464242826880013   0.464242826880013   0.408248290463863   0.303012985114696   0.161229841765317;
507        0.303012985114696   0.464242826880013   0.408248290463863   0.161229841765317  -0.161229841765317  -0.408248290463863  -0.464242826880013  -0.303012985114696;
508        0.408248290463863   0.408248290463863                   0  -0.408248290463863  -0.408248290463863                   0   0.408248290463863   0.408248290463863;
509        0.464242826880013   0.161229841765317  -0.408248290463863  -0.303012985114696   0.303012985114696   0.408248290463863  -0.161229841765317  -0.464242826880013;
510        0.464242826880013  -0.161229841765317  -0.408248290463863   0.303012985114696   0.303012985114696  -0.408248290463863  -0.161229841765317   0.464242826880013;
511        0.408248290463863  -0.408248290463863                   0   0.408248290463863  -0.408248290463863                   0   0.408248290463863  -0.408248290463863;
512        0.303012985114696  -0.464242826880013   0.408248290463863  -0.161229841765317  -0.161229841765317   0.408248290463863  -0.464242826880013   0.303012985114696;
513        0.161229841765317  -0.303012985114696   0.408248290463863  -0.464242826880013   0.464242826880013  -0.408248290463863   0.303012985114696  -0.161229841765317];
514 elseif (N == 7) & strcmp(transform_type, 'dct')==1 % hardcoded transform so that the signal processing toolbox is not needed to generate it
515     Tforward =[ 0.377964473009227   0.377964473009227   0.377964473009227   0.377964473009227   0.377964473009227   0.377964473009227   0.377964473009227;
516        0.521120889169602   0.417906505941275   0.231920613924330                   0  -0.231920613924330  -0.417906505941275  -0.521120889169602;
517        0.481588117120063   0.118942442321354  -0.333269317528993  -0.534522483824849  -0.333269317528993   0.118942442321354   0.481588117120063;
518        0.417906505941275  -0.231920613924330  -0.521120889169602                   0   0.521120889169602   0.231920613924330  -0.417906505941275;
519        0.333269317528993  -0.481588117120063  -0.118942442321354   0.534522483824849  -0.118942442321354  -0.481588117120063   0.333269317528993;
520        0.231920613924330  -0.521120889169602   0.417906505941275                   0  -0.417906505941275   0.521120889169602  -0.231920613924330;
521        0.118942442321354  -0.333269317528993   0.481588117120063  -0.534522483824849   0.481588117120063  -0.333269317528993   0.118942442321354];   
522 elseif strcmp(transform_type, 'dct') == 1,
523     Tforward    = dct(eye(N));
524 elseif strcmp(transform_type, 'dst') == 1,
525     Tforward    = dst(eye(N));
526 elseif strcmp(transform_type, 'DCrand') == 1,
527     x = randn(N); x(1:end,1) = 1; [Q,R] = qr(x); 
528     if (Q(1) < 0), 
529         Q = -Q; 
530     end;
531     Tforward = Q';
532 else %% a wavelet decomposition supported by 'wavedec'
533     %%% Set periodic boundary conditions, to preserve bi-orthogonality
534     dwtmode('per','nodisp');  
535     
536     Tforward = zeros(N,N);
537     for i = 1:N
538         Tforward(:,i)=wavedec(circshift([1 zeros(1,N-1)],[dec_levels i-1]), log2(N), transform_type);  %% construct transform matrix
539     end
540 end
541
542 %%% Normalize the basis elements
543 Tforward = (Tforward' * diag(sqrt(1./sum(Tforward.^2,2))))'; 
544
545 %%% Compute the inverse transform matrix
546 Tinverse = inv(Tforward);
547
548 return;