ivan@153: %%  Dictionary Learning for Image Denoising - KSVD vs Recursive Least Squares
ivan@153: %
ivan@153: %   This file contains an example of how SMALLbox can be used to test different
ivan@153: %   dictionary learning techniques in Image Denoising problem.
ivan@153: %   It calls generateImageDenoiseProblem that will let you to choose image,
ivan@153: %   add noise and use noisy image to generate training set for dictionary
ivan@153: %   learning.
ivan@153: %   Two dictionary learning techniques were compared:
ivan@153: %   -   KSVD - M. Elad, R. Rubinstein, and M. Zibulevsky, "Efficient
ivan@153: %              Implementation of the K-SVD Algorithm using Batch Orthogonal
ivan@153: %              Matching Pursuit", Technical Report - CS, Technion, April 2008.
ivan@153: %   -   RLS-DLA - Skretting, K.; Engan, K.; , "Recursive Least Squares
ivan@153: %       Dictionary Learning Algorithm," Signal Processing, IEEE Transactions on,
ivan@153: %       vol.58, no.4, pp.2121-2130, April 2010
ivan@153: %
ivan@153: 
ivan@153: 
ivan@153: %   Centre for Digital Music, Queen Mary, University of London.
ivan@153: %   This file copyright 2011 Ivan Damnjanovic.
ivan@153: %
ivan@153: %   This program is free software; you can redistribute it and/or
ivan@153: %   modify it under the terms of the GNU General Public License as
ivan@153: %   published by the Free Software Foundation; either version 2 of the
ivan@153: %   License, or (at your option) any later version.  See the file
ivan@153: %   COPYING included with this distribution for more information.
ivan@153: %   
ivan@153: %%
ivan@153: 
ivan@153: 
ivan@153: 
ivan@153: %   If you want to load the image outside of generateImageDenoiseProblem
ivan@153: %   function uncomment following lines. This can be useful if you want to
ivan@153: %   denoise more then one image for example.
ivan@153: %   Here we are loading test_image.mat that contains structure with 5 images : lena,
ivan@153: %   barbara,boat, house and peppers.
ivan@153: clear;
ivan@153: TMPpath=pwd;
ivan@153: FS=filesep;
luis@186: [pathstr1, name, ext] = fileparts(which('SMALLboxSetup.m'));
ivan@153: cd([pathstr1,FS,'data',FS,'images']);
ivan@153: load('test_image.mat');
ivan@153: cd(TMPpath);
ivan@153: 
ivan@153: %   Deffining the noise levels that we want to test
ivan@153: 
ivan@153: noise_level=[10 20 25 50 100];
ivan@153: 
ivan@153: %   Here we loop through different noise levels and images 
ivan@153: 
ivan@153: for noise_ind=2:2
ivan@153: for im_num=2:2
ivan@153: 
ivan@153: % Defining Image Denoising Problem as Dictionary Learning
ivan@153: % Problem. As an input we set the number of training patches.
ivan@153: 
ivan@153: SMALL.Problem = generateImageDenoiseProblem(test_image(im_num).i, 40000, '',256, noise_level(noise_ind));
ivan@153: SMALL.Problem.name=int2str(im_num);
ivan@153: 
ivan@153: Edata=sqrt(prod(SMALL.Problem.blocksize)) * SMALL.Problem.sigma * SMALL.Problem.gain;
ivan@153: maxatoms = floor(prod(SMALL.Problem.blocksize)/2);
ivan@153: 
ivan@153: 
ivan@153: %%
ivan@153: %   Use KSVD Dictionary Learning Algorithm to Learn overcomplete dictionary
ivan@153: %   Boris Mailhe ksvd update implentation omp is the same as with Rubinstein
ivan@153: %   implementation
ivan@153: 
ivan@153: 
ivan@153: %   Initialising solver structure
ivan@153: %   Setting solver structure fields (toolbox, name, param, solution,
ivan@153: %   reconstructed and time) to zero values
ivan@153: 
ivan@153: SMALL.solver(1)=SMALL_init_solver;
ivan@153: 
ivan@153: % Defining the parameters needed for image denoising
ivan@153: 
ivan@153: SMALL.solver(1).toolbox='ompbox';
ivan@153: SMALL.solver(1).name='omp2';
ivan@153: SMALL.solver(1).param=struct(...
ivan@153:     'epsilon',Edata,...
ivan@153:     'maxatoms', maxatoms); 
ivan@153: 
ivan@153: %   Initialising Dictionary structure
ivan@153: %   Setting Dictionary structure fields (toolbox, name, param, D and time)
ivan@153: %   to zero values
ivan@153: 
ivan@153: SMALL.DL(1)=SMALL_init_DL('TwoStepDL', 'KSVD', '', 1);
ivan@153: 
ivan@153: 
ivan@153: %   Defining the parameters for KSVD
ivan@153: %   In this example we are learning 256 atoms in 20 iterations, so that
ivan@153: %   every patch in the training set can be represented with target error in
ivan@153: %   L2-norm (EData)
ivan@153: %   Type help ksvd in MATLAB prompt for more options.
ivan@153: 
ivan@153: 
ivan@153: SMALL.DL(1).param=struct(...
ivan@153:     'solver', SMALL.solver(1),...
ivan@153:     'initdict', SMALL.Problem.initdict,...
ivan@153:     'dictsize', SMALL.Problem.p,...
ivan@153:     'iternum', 20,...
ivan@153:     'show_dict', 1);
ivan@153: 
ivan@153: %   Learn the dictionary
ivan@153: 
ivan@153: SMALL.DL(1) = SMALL_learn(SMALL.Problem, SMALL.DL(1));
ivan@153: 
ivan@153: %   Set SMALL.Problem.A dictionary
ivan@153: %   (backward compatiblity with SPARCO: solver structure communicate
ivan@153: %   only with Problem structure, ie no direct communication between DL and
ivan@153: %   solver structures)
ivan@153: 
ivan@153: SMALL.Problem.A = SMALL.DL(1).D;
ivan@161: SMALL.Problem.reconstruct = @(x) ImageDenoise_reconstruct(x, SMALL.Problem);
ivan@153: 
ivan@153: %   Denoising the image - find the sparse solution in the learned
ivan@153: %   dictionary for all patches in the image and the end it uses
ivan@153: %   reconstruction function to reconstruct the patches and put them into a
ivan@153: %   denoised image
ivan@153: 
ivan@153: SMALL.solver(1)=SMALL_solve(SMALL.Problem, SMALL.solver(1));
ivan@153: 
ivan@153: %%
ivan@153: %   Use MOD Dictionary Learning Algorithm to Learn overcomplete dictionary
ivan@153: %   Boris Mailhe MOD update implentation omp is the Ron Rubinstein
ivan@153: %   implementation
ivan@153: 
ivan@153: 
ivan@153: %   Initialising solver structure
ivan@153: %   Setting solver structure fields (toolbox, name, param, solution,
ivan@153: %   reconstructed and time) to zero values
ivan@153: 
ivan@153: SMALL.solver(2)=SMALL_init_solver;
ivan@153: 
ivan@153: % Defining the parameters needed for image denoising
ivan@153: 
ivan@153: SMALL.solver(2).toolbox='ompbox';
ivan@153: SMALL.solver(2).name='omp2';
ivan@153: SMALL.solver(2).param=struct(...
ivan@153:     'epsilon',Edata,...
ivan@153:     'maxatoms', maxatoms); 
ivan@153: 
ivan@153: %   Initialising Dictionary structure
ivan@153: %   Setting Dictionary structure fields (toolbox, name, param, D and time)
ivan@153: %   to zero values
ivan@153: 
ivan@153: SMALL.DL(2)=SMALL_init_DL('TwoStepDL', 'MOD', '', 1);
ivan@153: 
ivan@153: 
ivan@153: %   Defining the parameters for MOD
ivan@153: %   In this example we are learning 256 atoms in 20 iterations, so that
ivan@153: %   every patch in the training set can be represented with target error in
ivan@153: %   L2-norm (EData)
ivan@153: %   Type help ksvd in MATLAB prompt for more options
ivan@153: 
ivan@153: SMALL.DL(2).param=struct(...
ivan@153:     'solver', SMALL.solver(2),...
ivan@153:     'initdict', SMALL.Problem.initdict,...
ivan@153:     'dictsize', SMALL.Problem.p,...
ivan@153:     'iternum', 20,...
ivan@153:     'show_dict', 1);
ivan@153: 
ivan@153: %   Learn the dictionary
ivan@153: 
ivan@153: SMALL.DL(2) = SMALL_learn(SMALL.Problem, SMALL.DL(2));
ivan@153: 
ivan@153: %   Set SMALL.Problem.A dictionary
ivan@153: %   (backward compatiblity with SPARCO: solver structure communicate
ivan@153: %   only with Problem structure, ie no direct communication between DL and
ivan@153: %   solver structures)
ivan@153: 
ivan@153: SMALL.Problem.A = SMALL.DL(2).D;
ivan@161: SMALL.Problem.reconstruct = @(x) ImageDenoise_reconstruct(x, SMALL.Problem);
ivan@153: 
ivan@153: %   Denoising the image - find the sparse solution in the learned
ivan@153: %   dictionary for all patches in the image and the end it uses
ivan@153: %   reconstruction function to reconstruct the patches and put them into a
ivan@153: %   denoised image
ivan@153: 
ivan@153: SMALL.solver(2)=SMALL_solve(SMALL.Problem, SMALL.solver(2));
ivan@153: %%
ivan@153: %   Use OLS Dictionary Learning Algorithm to Learn overcomplete dictionary
ivan@153: %   Boris Mailhe ksvd update implentation omp is the Ron Rubinstein
ivan@153: %   implementation
ivan@153: 
ivan@153: 
ivan@153: %   Initialising solver structure
ivan@153: %   Setting solver structure fields (toolbox, name, param, solution,
ivan@153: %   reconstructed and time) to zero values
ivan@153: 
ivan@153: SMALL.solver(3)=SMALL_init_solver;
ivan@153: 
ivan@153: % Defining the parameters needed for image denoising
ivan@153: 
ivan@153: SMALL.solver(3).toolbox='ompbox';
ivan@153: SMALL.solver(3).name='omp2';
ivan@153: SMALL.solver(3).param=struct(...
ivan@153:     'epsilon',Edata,...
ivan@153:     'maxatoms', maxatoms); 
ivan@153: 
ivan@153: %   Initialising Dictionary structure
ivan@153: %   Setting Dictionary structure fields (toolbox, name, param, D and time)
ivan@153: %   to zero values
ivan@153: 
ivan@153: SMALL.DL(3)=SMALL_init_DL('TwoStepDL', 'ols', '', 1);
ivan@153: 
ivan@153: 
ivan@153: %   Defining the parameters for KSVD
ivan@153: %   In this example we are learning 256 atoms in 20 iterations, so that
ivan@153: %   every patch in the training set can be represented with target error in
ivan@153: %   L2-norm (EData)
ivan@153: %   Type help ksvd in MATLAB prompt for more options.
ivan@153: 
ivan@153: 
ivan@153: SMALL.DL(3).param=struct(...
ivan@153:     'solver', SMALL.solver(3),...
ivan@153:     'initdict', SMALL.Problem.initdict,...
ivan@153:     'dictsize', SMALL.Problem.p,...
ivan@153:     'iternum', 20,...
ivan@153:     'learningRate', 0.1,...
ivan@153:     'show_dict', 1);
ivan@153: 
ivan@153: %   Learn the dictionary
ivan@153: 
ivan@153: SMALL.DL(3) = SMALL_learn(SMALL.Problem, SMALL.DL(3));
ivan@153: 
ivan@153: %   Set SMALL.Problem.A dictionary
ivan@153: %   (backward compatiblity with SPARCO: solver structure communicate
ivan@153: %   only with Problem structure, ie no direct communication between DL and
ivan@153: %   solver structures)
ivan@153: 
ivan@153: SMALL.Problem.A = SMALL.DL(3).D;
ivan@161: SMALL.Problem.reconstruct = @(x) ImageDenoise_reconstruct(x, SMALL.Problem);
ivan@153: 
ivan@153: %   Denoising the image - find the sparse solution in the learned
ivan@153: %   dictionary for all patches in the image and the end it uses
ivan@153: %   reconstruction function to reconstruct the patches and put them into a
ivan@153: %   denoised image
ivan@153: 
ivan@153: SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
ivan@153: %%
ivan@153: %   Use Mailhe Dictionary Learning Algorithm to Learn overcomplete dictionary
ivan@153: %   Boris Mailhe ksvd update implentation omp is the Ron Rubinstein
ivan@153: %   implementation
ivan@153: 
ivan@153: 
ivan@153: %   Initialising solver structure
ivan@153: %   Setting solver structure fields (toolbox, name, param, solution,
ivan@153: %   reconstructed and time) to zero values
ivan@153: 
ivan@153: SMALL.solver(4)=SMALL_init_solver;
ivan@153: 
ivan@153: % Defining the parameters needed for image denoising
ivan@153: 
ivan@153: SMALL.solver(4).toolbox='ompbox';
ivan@153: SMALL.solver(4).name='omp2';
ivan@153: SMALL.solver(4).param=struct(...
ivan@153:     'epsilon',Edata,...
ivan@153:     'maxatoms', maxatoms); 
ivan@153: 
ivan@153: %   Initialising Dictionary structure
ivan@153: %   Setting Dictionary structure fields (toolbox, name, param, D and time)
ivan@153: %   to zero values
ivan@153: 
bmailhe@212: SMALL.DL(4)=SMALL_init_DL('TwoStepDL', 'opt', '', 1);
ivan@153: 
ivan@153: 
ivan@153: %   Defining the parameters for KSVD
ivan@153: %   In this example we are learning 256 atoms in 20 iterations, so that
ivan@153: %   every patch in the training set can be represented with target error in
ivan@153: %   L2-norm (EData)
ivan@153: %   Type help ksvd in MATLAB prompt for more options.
ivan@153: 
ivan@153: 
ivan@153: SMALL.DL(4).param=struct(...
ivan@153:     'solver', SMALL.solver(4),...
ivan@153:     'initdict', SMALL.Problem.initdict,...
ivan@153:     'dictsize', SMALL.Problem.p,...
ivan@153:     'iternum', 20,...
ivan@153:     'learningRate', 2,...
ivan@153:     'show_dict', 1);
ivan@153: 
ivan@153: %   Learn the dictionary
ivan@153: 
ivan@153: SMALL.DL(4) = SMALL_learn(SMALL.Problem, SMALL.DL(4));
ivan@153: 
ivan@153: %   Set SMALL.Problem.A dictionary
ivan@153: %   (backward compatiblity with SPARCO: solver structure communicate
ivan@153: %   only with Problem structure, ie no direct communication between DL and
ivan@153: %   solver structures)
ivan@153: 
ivan@153: SMALL.Problem.A = SMALL.DL(4).D;
ivan@161: SMALL.Problem.reconstruct = @(x) ImageDenoise_reconstruct(x, SMALL.Problem);
ivan@153: 
ivan@153: %   Denoising the image - find the sparse solution in the learned
ivan@153: %   dictionary for all patches in the image and the end it uses
ivan@153: %   reconstruction function to reconstruct the patches and put them into a
ivan@153: %   denoised image
ivan@153: 
ivan@153: SMALL.solver(4)=SMALL_solve(SMALL.Problem, SMALL.solver(4));
ivan@153: 
ivan@153: %% show results %%
ivan@153: 
ivan@153: SMALL_ImgDeNoiseResult(SMALL);
ivan@153: 
ivan@153: %clear SMALL;
ivan@153: end
ivan@153: end
ivan@153: