ivan@128: %% Dictionary Learning for Image Denoising - KSVD vs KSVDS vs SPAMS
ivan@128: %
ivan@128: % *WARNING!* You should have SPAMS in your search path in order for this
ivan@128: % script to work.Due to licensing issues SPAMS can not be automatically
ivan@128: % provided in SMALLbox (http://www.di.ens.fr/willow/SPAMS/downloads.html).
ivan@128: %
idamnjanovic@6: % This file contains an example of how SMALLbox can be used to test different
idamnjanovic@6: % dictionary learning techniques in Image Denoising problem.
idamnjanovic@6: % It calls generateImageDenoiseProblem that will let you to choose image,
idamnjanovic@6: % add noise and use noisy image to generate training set for dictionary
idamnjanovic@6: % learning.
idamnjanovic@6: % Three dictionary learning techniques were compared:
idamnjanovic@6: % - KSVD - M. Elad, R. Rubinstein, and M. Zibulevsky, "Efficient
idamnjanovic@6: % Implementation of the K-SVD Algorithm using Batch Orthogonal
idamnjanovic@6: % Matching Pursuit", Technical Report - CS, Technion, April 2008.
idamnjanovic@6: % - KSVDS - R. Rubinstein, M. Zibulevsky, and M. Elad, "Learning Sparse
idamnjanovic@6: % Dictionaries for Sparse Signal Approximation", Technical
idamnjanovic@6: % Report - CS, Technion, June 2009.
idamnjanovic@6: % - SPAMS - J. Mairal, F. Bach, J. Ponce and G. Sapiro. Online
idamnjanovic@6: % Dictionary Learning for Sparse Coding. International
idamnjanovic@6: % Conference on Machine Learning,Montreal, Canada, 2009
idamnjanovic@6: %
ivan@107:
ivan@107: %
ivan@107: % Centre for Digital Music, Queen Mary, University of London.
ivan@107: % This file copyright 2009 Ivan Damnjanovic.
ivan@107: %
ivan@107: % This program is free software; you can redistribute it and/or
ivan@107: % modify it under the terms of the GNU General Public License as
ivan@107: % published by the Free Software Foundation; either version 2 of the
ivan@107: % License, or (at your option) any later version. See the file
ivan@107: % COPYING included with this distribution for more information.
idamnjanovic@6: %
idamnjanovic@6: %%
idamnjanovic@6:
idamnjanovic@6: clear;
idamnjanovic@6:
idamnjanovic@6: % If you want to load the image outside of generateImageDenoiseProblem
idamnjanovic@6: % function uncomment following lines. This can be useful if you want to
idamnjanovic@6: % denoise more then one image for example.
idamnjanovic@6:
idamnjanovic@6: % TMPpath=pwd;
idamnjanovic@6: % FS=filesep;
idamnjanovic@6: % [pathstr1, name, ext, versn] = fileparts(which('SMALLboxSetup.m'));
idamnjanovic@6: % cd([pathstr1,FS,'data',FS,'images']);
idamnjanovic@6: % [filename,pathname] = uigetfile({'*.png;'},'Select a file containin pre-calculated notes');
idamnjanovic@6: % [pathstr, name, ext, versn] = fileparts(filename);
idamnjanovic@6: % test_image = imread(filename);
idamnjanovic@6: % test_image = double(test_image);
idamnjanovic@6: % cd(TMPpath);
idamnjanovic@6: % SMALL.Problem.name=name;
idamnjanovic@6:
idamnjanovic@6:
idamnjanovic@6: % Defining Image Denoising Problem as Dictionary Learning
idamnjanovic@6: % Problem. As an input we set the number of training patches.
idamnjanovic@6:
idamnjanovic@6: SMALL.Problem = generateImageDenoiseProblem('', 40000);
idamnjanovic@6:
idamnjanovic@6:
idamnjanovic@6: %%
idamnjanovic@6: % Use KSVD Dictionary Learning Algorithm to Learn overcomplete dictionary
idamnjanovic@6:
idamnjanovic@6: % Initialising Dictionary structure
idamnjanovic@6: % Setting Dictionary structure fields (toolbox, name, param, D and time)
idamnjanovic@6: % to zero values
idamnjanovic@6:
idamnjanovic@6: SMALL.DL(1)=SMALL_init_DL();
idamnjanovic@6:
idamnjanovic@6: % Defining the parameters needed for dictionary learning
idamnjanovic@6:
idamnjanovic@6: SMALL.DL(1).toolbox = 'KSVD';
idamnjanovic@6: SMALL.DL(1).name = 'ksvd';
idamnjanovic@6:
idamnjanovic@6: % Defining the parameters for KSVD
idamnjanovic@6: % In this example we are learning 256 atoms in 20 iterations, so that
idamnjanovic@6: % every patch in the training set can be represented with target error in
idamnjanovic@6: % L2-norm (EData)
idamnjanovic@6: % Type help ksvd in MATLAB prompt for more options.
idamnjanovic@6:
idamnjanovic@6: Edata=sqrt(prod(SMALL.Problem.blocksize)) * SMALL.Problem.sigma * SMALL.Problem.gain;
ivan@107: maxatoms = floor(prod(SMALL.Problem.blocksize)/2);
ivan@107:
idamnjanovic@6: SMALL.DL(1).param=struct(...
idamnjanovic@6: 'Edata', Edata,...
idamnjanovic@6: 'initdict', SMALL.Problem.initdict,...
idamnjanovic@6: 'dictsize', SMALL.Problem.p,...
idamnjanovic@6: 'iternum', 20,...
idamnjanovic@6: 'memusage', 'high');
idamnjanovic@6:
idamnjanovic@6: % Learn the dictionary
idamnjanovic@6:
idamnjanovic@6: SMALL.DL(1) = SMALL_learn(SMALL.Problem, SMALL.DL(1));
idamnjanovic@6:
idamnjanovic@6: % Set SMALL.Problem.A dictionary
idamnjanovic@6: % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@6: % only with Problem structure, ie no direct communication between DL and
idamnjanovic@6: % solver structures)
idamnjanovic@6:
idamnjanovic@6: SMALL.Problem.A = SMALL.DL(1).D;
ivan@107: SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem);
idamnjanovic@6:
idamnjanovic@6: %%
idamnjanovic@6: % Initialising solver structure
idamnjanovic@6: % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@6: % reconstructed and time) to zero values
idamnjanovic@6:
idamnjanovic@6: SMALL.solver(1)=SMALL_init_solver;
idamnjanovic@6:
idamnjanovic@6: % Defining the parameters needed for image denoising
idamnjanovic@6:
idamnjanovic@6: SMALL.solver(1).toolbox='ompbox';
ivan@107: SMALL.solver(1).name='omp2';
ivan@107: SMALL.solver(1).param=struct(...
ivan@107: 'epsilon',Edata,...
ivan@107: 'maxatoms', maxatoms);
idamnjanovic@6:
ivan@107: % Denoising the image - find the sparse solution in the learned
ivan@107: % dictionary for all patches in the image and the end it uses
ivan@107: % reconstruction function to reconstruct the patches and put them into a
ivan@107: % denoised image
idamnjanovic@6:
ivan@107: SMALL.solver(1)=SMALL_solve(SMALL.Problem, SMALL.solver(1));
ivan@107:
ivan@107: % Show PSNR after reconstruction
ivan@107:
ivan@107: SMALL.solver(1).reconstructed.psnr
idamnjanovic@6:
idamnjanovic@6: %%
idamnjanovic@6: % Use KSVDS Dictionary Learning Algorithm to denoise image
idamnjanovic@6:
idamnjanovic@6: % Initialising solver structure
idamnjanovic@6: % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@6: % reconstructed and time) to zero values
idamnjanovic@6:
idamnjanovic@6: SMALL.DL(2)=SMALL_init_DL();
idamnjanovic@6:
idamnjanovic@6: % Defining the parameters needed for dictionary learning
idamnjanovic@6:
idamnjanovic@6: SMALL.DL(2).toolbox = 'KSVDS';
idamnjanovic@6: SMALL.DL(2).name = 'ksvds';
idamnjanovic@6:
idamnjanovic@6: % Defining the parameters for KSVDS
idamnjanovic@6: % In this example we are learning 256 atoms in 20 iterations, so that
idamnjanovic@6: % every patch in the training set can be represented with target error in
idamnjanovic@6: % L2-norm (EDataS). We also impose "double sparsity" - dictionary itself
idamnjanovic@6: % has to be sparse in the given base dictionary (Tdict - number of
idamnjanovic@6: % nonzero elements per atom).
idamnjanovic@6: % Type help ksvds in MATLAB prompt for more options.
idamnjanovic@6:
idamnjanovic@6: EdataS=sqrt(prod(SMALL.Problem.blocksize)) * SMALL.Problem.sigma * SMALL.Problem.gain;
idamnjanovic@6: SMALL.DL(2).param=struct(...
idamnjanovic@6: 'Edata', EdataS, ...
idamnjanovic@6: 'Tdict', 6,...
idamnjanovic@6: 'stepsize', 1,...
idamnjanovic@6: 'dictsize', SMALL.Problem.p,...
idamnjanovic@6: 'iternum', 20,...
idamnjanovic@6: 'memusage', 'high');
idamnjanovic@6: SMALL.DL(2).param.initA = speye(SMALL.Problem.p);
idamnjanovic@6: SMALL.DL(2).param.basedict{1} = odctdict(8,16);
idamnjanovic@6: SMALL.DL(2).param.basedict{2} = odctdict(8,16);
idamnjanovic@6:
idamnjanovic@6: % Learn the dictionary
idamnjanovic@6:
idamnjanovic@6: SMALL.DL(2) = SMALL_learn(SMALL.Problem, SMALL.DL(2));
idamnjanovic@6:
idamnjanovic@6: % Set SMALL.Problem.A dictionary and SMALL.Problem.basedictionary
idamnjanovic@6: % (backward compatiblity with SPARCO: solver structure communicate
idamnjanovic@6: % only with Problem structure, ie no direct communication between DL and
idamnjanovic@6: % solver structures)
idamnjanovic@6:
idamnjanovic@6: SMALL.Problem.A = SMALL.DL(2).D;
idamnjanovic@6: SMALL.Problem.basedict{1} = SMALL.DL(2).param.basedict{1};
idamnjanovic@6: SMALL.Problem.basedict{2} = SMALL.DL(2).param.basedict{2};
idamnjanovic@6:
ivan@107: % Setting up reconstruction function
ivan@107:
ivan@107: SparseDict=1;
ivan@107: SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem, SparseDict);
ivan@107:
idamnjanovic@6: % Initialising solver structure
idamnjanovic@6: % Setting solver structure fields (toolbox, name, param, solution,
idamnjanovic@6: % reconstructed and time) to zero values
idamnjanovic@6:
idamnjanovic@6: SMALL.solver(2)=SMALL_init_solver;
idamnjanovic@6:
idamnjanovic@6: % Defining the parameters needed for image denoising
idamnjanovic@6:
idamnjanovic@6: SMALL.solver(2).toolbox='ompsbox';
ivan@107: SMALL.solver(2).name='omps2';
ivan@107: SMALL.solver(2).param=struct(...
ivan@107: 'epsilon',Edata,...
ivan@107: 'maxatoms', maxatoms);
idamnjanovic@6:
ivan@107: % Denoising the image - find the sparse solution in the learned
ivan@107: % dictionary for all patches in the image and the end it uses
ivan@107: % reconstruction function to reconstruct the patches and put them into a
ivan@107: % denoised image
idamnjanovic@6:
ivan@107: SMALL.solver(2)=SMALL_solve(SMALL.Problem, SMALL.solver(2));
idamnjanovic@6:
ivan@107: %%
ivan@107: % Use SPAMS Online Dictionary Learning Algorithm
ivan@107: % to Learn overcomplete dictionary (Julien Mairal 2009)
ivan@107: % (If you have not installed SPAMS please comment the following two cells)
ivan@107:
ivan@107: % Initialising Dictionary structure
ivan@107: % Setting Dictionary structure fields (toolbox, name, param, D and time)
ivan@107: % to zero values
ivan@107:
ivan@107: SMALL.DL(3)=SMALL_init_DL();
ivan@107:
ivan@107: % Defining fields needed for dictionary learning
ivan@107:
ivan@107: SMALL.DL(3).toolbox = 'SPAMS';
ivan@107: SMALL.DL(3).name = 'mexTrainDL';
ivan@107:
ivan@107: % Type 'help mexTrainDL in MATLAB prompt for explanation of parameters.
ivan@107:
ivan@107: SMALL.DL(3).param=struct(...
ivan@107: 'D', SMALL.Problem.initdict,...
ivan@107: 'K', SMALL.Problem.p,...
ivan@107: 'lambda', 2,...
ivan@107: 'iter', 200,...
ivan@107: 'mode', 3, ...
ivan@107: 'modeD', 0);
ivan@107:
ivan@107: % Learn the dictionary
ivan@107:
ivan@107: SMALL.DL(3) = SMALL_learn(SMALL.Problem, SMALL.DL(3));
ivan@107:
ivan@107: % Set SMALL.Problem.A dictionary
ivan@107: % (backward compatiblity with SPARCO: solver structure communicate
ivan@107: % only with Problem structure, ie no direct communication between DL and
ivan@107: % solver structures)
ivan@107:
ivan@107: SMALL.Problem.A = SMALL.DL(3).D;
ivan@107:
ivan@107: % Setting up reconstruction function
ivan@107:
ivan@107: SMALL.Problem.reconstruct = @(x) ImgDenoise_reconstruct(x, SMALL.Problem);
ivan@107:
ivan@107: % Initialising solver structure
ivan@107: % Setting solver structure fields (toolbox, name, param, solution,
ivan@107: % reconstructed and time) to zero values
ivan@107:
ivan@107: SMALL.solver(3)=SMALL_init_solver;
ivan@107:
ivan@107: % Defining the parameters needed for image denoising
ivan@107:
ivan@107: SMALL.solver(3).toolbox='ompbox';
ivan@107: SMALL.solver(3).name='omp2';
ivan@107: SMALL.solver(3).param=struct(...
ivan@107: 'epsilon',Edata,...
ivan@107: 'maxatoms', maxatoms);
ivan@107:
ivan@107: % Denoising the image - find the sparse solution in the learned
ivan@107: % dictionary for all patches in the image and the end it uses
ivan@107: % reconstruction function to reconstruct the patches and put them into a
ivan@107: % denoised image
ivan@107:
ivan@107: SMALL.solver(3)=SMALL_solve(SMALL.Problem, SMALL.solver(3));
idamnjanovic@6:
idamnjanovic@6: %%
idamnjanovic@6: % Plot results and save midi files
idamnjanovic@6:
idamnjanovic@6: % show results %
idamnjanovic@6:
idamnjanovic@6: SMALL_ImgDeNoiseResult(SMALL);