Mercurial > hg > smallbox

comparison examples/SMALL_test_coherence.m @ 169:290cca7d3469 danieleb

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Added dictionary decorrelation functions and test script for ICASSP paper.
author Daniele Barchiesi <daniele.barchiesi@eecs.qmul.ac.uk>
date Thu, 29 Sep 2011 09:46:52 +0100
parents 8324c7ea6602
children
comparison
equal deleted inserted replaced
168:ff866a412be5 169:290cca7d3469
16 percActiveAtoms = 5; %percentage of active atoms 16 percActiveAtoms = 5; %percentage of active atoms
17 17
18 % Dependent parameters 18 % Dependent parameters
19 nActiveAtoms = fix(blockSize/100*percActiveAtoms); %number of active atoms 19 nActiveAtoms = fix(blockSize/100*percActiveAtoms); %number of active atoms
20 epsilon = 1/sigma; %error constraint for sparse representation step (corresponds to noise applied to signals) 20 epsilon = 1/sigma; %error constraint for sparse representation step (corresponds to noise applied to signals)
21 %minCoherence = sqrt((dictSize-blockSize)/(blockSize*(dictSize-1))); %target coherence (based on coherence lower bound) 21 minMu = sqrt((dictSize-blockSize)/(blockSize*(dictSize-1))); %target coherence (based on coherence lower bound)
22 minCoherence = 0.4; %target coherence 22 minCoherence = [0.1:0.1:1];
23 23
24 % Initial dictionaries 24 % Initial dictionaries
25 dctDict = dictionary('dct',blockSize,dictSize); 25 dctDict = dictionary('dct',blockSize,dictSize);
26 dctDict = dctDict.phi; 26 dctDict = dctDict.phi;
27 gaborParam = struct('N',blockSize,'redundancyFactor',2,'wd',@rectwin); 27 gaborParam = struct('N',blockSize,'redundancyFactor',2,'wd',@rectwin);
28 gaborDict = Gabor_Dictionary(gaborParam); 28 gaborDict = Gabor_Dictionary(gaborParam);
29 29
30 %% Generate audio denoising problem with low noise (audio representation) 30 %% Generate audio denoising problem with low noise (audio representation)
31 SMALL.Problem = generateAudioDenoiseProblem(signal.s,[],blockSize,... 31 SMALL.Problem = generateAudioDenoiseProblem(signal.s,[],blockSize,...
32 dictSize,overlap,sigma); % generate representation problem 32 dictSize,overlap,sigma); % generate representation problem
33 SMALL.Problem.b1 = SMALL.Problem.b; % copy signals from training set b to test set b1 (needed for later functions) 33 SMALL.Problem.b1 = SMALL.Problem.b; % copy signals from training set b to test set b1 (needed for later functions)
34 34
35 % omp2 sparse representation solver 35 % omp2 sparse representation solver
36 ompParam = struct('X',SMALL.Problem.b,'epsilon',epsilon,'maxatoms',nActiveAtoms); %parameters 36 ompParam = struct('X',SMALL.Problem.b,'epsilon',epsilon,'maxatoms',nActiveAtoms); %parameters
37 solver = SMALL_init_solver('ompbox','omp2',ompParam,false); %solver structure 37 solver = SMALL_init_solver('ompbox','omp2',ompParam,false); %solver structure
38 38
39 39
40 %% Test ksvd dictionary update 40 %% Test
41 name = dicUpdate{1}; %use ksvd update 41 name = dicUpdate{1}; %use ksvd update
42 SMALL.DL(1:9) = SMALL_init_DL(toolbox,name); %create dictionary learning structures 42 SMALL.DL(1:36) = SMALL_init_DL(toolbox,name); %create dictionary learning structures
43 43
44 % learn with random initialisation and no decorrelation 44 % learn with random initialisation and no decorrelation
45 SMALL.DL(1).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 45 SMALL.DL(1).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
46 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 46 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
47 'decFcn','none'); %parameters for the dictionary learning 47 'decFcn','none'); %parameters for the dictionary learning
48 SMALL.DL(1) = SMALL_learn(SMALL.Problem,SMALL.DL(1)); %learn dictionary 48 SMALL.DL(1) = SMALL_learn(SMALL.Problem,SMALL.DL(1)); %learn dictionary
49 %save('SMALL','SMALL');
49 50
50 % learn with random initialisation and mailhe decorrelation 51 % learn with random initialisation and mailhe decorrelation
51 SMALL.DL(2).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 52 for iMu=1:10
52 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 53 SMALL.DL(1+iMu).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
53 'decFcn','mailhe','coherence',minCoherence); %parameters for the dictionary learning 54 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
54 SMALL.DL(2) = SMALL_learn(SMALL.Problem,SMALL.DL(2)); %learn dictionary 55 'decFcn','mailhe','coherence',minCoherence(iMu)); %parameters for the dictionary learning
55 56 SMALL.DL(1+iMu) = SMALL_learn(SMALL.Problem,SMALL.DL(1+iMu)); %learn dictionary
56 % learn with random initialisation and tropp decorrelation 57 %save('SMALL','SMALL');
57 SMALL.DL(3).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 58 end
58 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
59 'decFcn','tropp','coherence',minCoherence); %parameters for the dictionary learning
60 SMALL.DL(3) = SMALL_learn(SMALL.Problem,SMALL.DL(3)); %learn dictionary
61
62 % Learn with dct initialisation and no decorrelation
63 SMALL.DL(4).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
64 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
65 'decFcn','none','initdict',dctDict); %parameters for the dictionary learning
66 SMALL.DL(4) = SMALL_learn(SMALL.Problem,SMALL.DL(4)); %learn dictionary
67
68 % learn with dct initialisation and mailhe decorrelation
69 SMALL.DL(5).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
70 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
71 'decFcn','mailhe','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning
72 SMALL.DL(5) = SMALL_learn(SMALL.Problem,SMALL.DL(5)); %learn dictionary
73
74 % learn with dct initialisation and tropp decorrelation
75 SMALL.DL(6).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
76 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
77 'decFcn','tropp','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning
78 SMALL.DL(6) = SMALL_learn(SMALL.Problem,SMALL.DL(6)); %learn dictionary
79
80 % Learn with gabor initialisation and no decorrelation
81 SMALL.DL(7).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
82 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
83 'decFcn','none','initdict',gaborDict); %parameters for the dictionary learning
84 SMALL.DL(7) = SMALL_learn(SMALL.Problem,SMALL.DL(7)); %learn dictionary
85
86 % learn with gabor initialisation and mailhe decorrelation
87 SMALL.DL(8).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
88 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
89 'decFcn','mailhe','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning
90 SMALL.DL(8) = SMALL_learn(SMALL.Problem,SMALL.DL(8)); %learn dictionary
91
92 % learn with gabor initialisation and tropp decorrelation
93 SMALL.DL(9).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
94 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
95 'decFcn','tropp','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning
96 SMALL.DL(9) = SMALL_learn(SMALL.Problem,SMALL.DL(9)); %learn dictionary
97
98 %% Test mailhe dictionary update
99 name = dicUpdate{2}; %use mailhe update
100 SMALL.DL(10:18) = SMALL_init_DL(toolbox,name); %create dictionary learning structure
101
102 % learn with random initialisation and no decorrelation
103 SMALL.DL(10).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
104 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
105 'decFcn','none'); %parameters for the dictionary learning
106 SMALL.DL(10) = SMALL_learn(SMALL.Problem,SMALL.DL(10)); %learn dictionary
107
108 % learn with random initialisation and mailhe decorrelation
109 SMALL.DL(11).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
110 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
111 'decFcn','mailhe','coherence',minCoherence); %parameters for the dictionary learning
112 SMALL.DL(11) = SMALL_learn(SMALL.Problem,SMALL.DL(11)); %learn dictionary
113 59
114 % learn with random initialisation and tropp decorrelation 60 % learn with random initialisation and tropp decorrelation
115 SMALL.DL(12).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 61 SMALL.DL(12).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
116 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 62 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
117 'decFcn','tropp','coherence',minCoherence); %parameters for the dictionary learning 63 'decFcn','tropp','coherence',minCoherence); %parameters for the dictionary learning
118 SMALL.DL(12) = SMALL_learn(SMALL.Problem,SMALL.DL(12)); %learn dictionary 64 SMALL.DL(12) = SMALL_learn(SMALL.Problem,SMALL.DL(12)); %learn dictionary
119 65
120 % Learn with dct initialisation and no decorrelation 66 % Learn with dct initialisation and no decorrelation
121 SMALL.DL(13).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 67 SMALL.DL(13).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
122 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 68 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
123 'decFcn','none','initdict',dctDict); %parameters for the dictionary learning 69 'decFcn','none','initdict',dctDict); %parameters for the dictionary learning
124 SMALL.DL(13) = SMALL_learn(SMALL.Problem,SMALL.DL(13)); %learn dictionary 70 SMALL.DL(13) = SMALL_learn(SMALL.Problem,SMALL.DL(13)); %learn dictionary
125 71
126 % learn with dct initialisation and mailhe decorrelation 72 % learn with dct initialisation and mailhe decorrelation
127 SMALL.DL(14).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 73 for iMu=1:10
128 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 74 SMALL.DL(13+iMu).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
129 'decFcn','mailhe','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning 75 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
130 SMALL.DL(14) = SMALL_learn(SMALL.Problem,SMALL.DL(14)); %learn dictionary 76 'decFcn','mailhe','coherence',minCoherence(iMu),'initdict',dctDict); %parameters for the dictionary learning
77 SMALL.DL(13+iMu) = SMALL_learn(SMALL.Problem,SMALL.DL(13+iMu)); %learn dictionary
78 %save('SMALL','SMALL');
79 end
131 80
132 % learn with dct initialisation and tropp decorrelation 81 % learn with dct initialisation and tropp decorrelation
133 SMALL.DL(15).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 82 SMALL.DL(24).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
134 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 83 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
135 'decFcn','tropp','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning 84 'decFcn','tropp','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning
136 SMALL.DL(15) = SMALL_learn(SMALL.Problem,SMALL.DL(15)); %learn dictionary 85 SMALL.DL(24) = SMALL_learn(SMALL.Problem,SMALL.DL(24)); %learn dictionary
86
137 87
138 % Learn with gabor initialisation and no decorrelation 88 % Learn with gabor initialisation and no decorrelation
139 SMALL.DL(16).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 89 SMALL.DL(25).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
140 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 90 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
141 'decFcn','none','initdict',gaborDict); %parameters for the dictionary learning 91 'decFcn','none','initdict',gaborDict); %parameters for the dictionary learning
142 SMALL.DL(16) = SMALL_learn(SMALL.Problem,SMALL.DL(16)); %learn dictionary 92 SMALL.DL(25) = SMALL_learn(SMALL.Problem,SMALL.DL(25)); %learn dictionary
143 93
144 % learn with gabor initialisation and mailhe decorrelation 94 % learn with gabor initialisation and mailhe decorrelation
145 SMALL.DL(17).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 95 for iMu=1:10
146 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 96 SMALL.DL(25+iMu).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
147 'decFcn','mailhe','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning 97 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
148 SMALL.DL(17) = SMALL_learn(SMALL.Problem,SMALL.DL(17)); %learn dictionary 98 'decFcn','mailhe','coherence',minCoherence(iMu),'initdict',gaborDict); %parameters for the dictionary learning
99 SMALL.DL(25+iMu) = SMALL_learn(SMALL.Problem,SMALL.DL(25+iMu)); %learn dictionary
100 %save('SMALL','SMALL');
101 end
149 102
150 % learn with gabor initialisation and tropp decorrelation 103 % learn with gabor initialisation and tropp decorrelation
151 SMALL.DL(18).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,... 104 SMALL.DL(36).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
152 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,... 105 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
153 'decFcn','tropp','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning 106 'decFcn','tropp','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning
154 SMALL.DL(18) = SMALL_learn(SMALL.Problem,SMALL.DL(18)); %learn dictionary 107 SMALL.DL(36) = SMALL_learn(SMALL.Problem,SMALL.DL(36)); %learn dictionary
155 108
156 %% Evaluate coherence and snr of representation for the various methods 109 %% Evaluate coherence and snr of representation for the various methods
157 sigNoiseRatio = zeros(18,1); 110 sigNoiseRatio = zeros(36,1);
158 mu = zeros(18,1); 111 mu = zeros(36,1);
159 for i=1:18 112 for i=1:36
160 SMALL.Problem.A = SMALL.DL(i).D; 113 SMALL.Problem.A = SMALL.DL(i).D;
161 tempSolver = SMALL_solve(SMALL.Problem,solver); 114 tempSolver = SMALL_solve(SMALL.Problem,solver);
162 sigNoiseRatio(i) = snr(SMALL.Problem.b,SMALL.DL(i).D*tempSolver.solution); 115 sigNoiseRatio(i) = snr(SMALL.Problem.b,SMALL.DL(i).D*tempSolver.solution);
163 dic(i) = dictionary(SMALL.DL(i).D); 116 dic(i) = dictionary(SMALL.DL(i).D);
164 mu(i) = dic(i).coherence; 117 mu(i) = dic(i).coherence;
165 end 118 end
166 119
167 %% Plot results 120 %% Plot results
168 minMu = sqrt((dictSize-blockSize)/(blockSize*(dictSize-1))); 121 minMu = sqrt((dictSize-blockSize)/(blockSize*(dictSize-1)));
169 maxSNR = max(sigNoiseRatio); 122
170 123 figure,
171 figure, subplot(1,2,1) 124 %subplot(3,1,1)
172 snrMat = buffer(sigNoiseRatio(1:9),3); 125 hold on, grid on
173 bar(snrMat'); 126 title('Data Initialisation')
174 title('Signal to noise ratio') 127 plot([1 1],[0 25],'k-');
175 xlabel('Initial dictionary') 128 plot(mu(1),sigNoiseRatio(1),'ks');
176 ylabel('SNR (dB)') 129 plot(mu(12),sigNoiseRatio(12),'kd');
177 set(gca,'XTickLabel',{'data','dct','gabor'}); 130 plot(mu(2:11),sigNoiseRatio(2:11),'k*-');
178 legend('none','Mailhe','Tropp') 131 plot([minMu minMu],[0 25],'k--')
179 grid on 132 set(gca,'YLim',[0 25],'XLim',[0 1.4]);
180 133 legend({'\mu_{max}','K-SVD','Grassmannian','INK-SVD','\mu_{min}'});
134 xlabel('\mu');
135 ylabel('SNR (dB)');
136
137 figure
138 %subplot(3,1,2)
139 hold on, grid on
140 title('DCT Initialisation')
141 plot([1 1],[0 25],'k-');
142 plot(mu(13),sigNoiseRatio(13),'ks');
143 plot(mu(24),sigNoiseRatio(24),'kd');
144 plot(mu(14:23),sigNoiseRatio(14:23),'k*-');
145 plot([minMu minMu],[0 25],'k--')
146 set(gca,'YLim',[0 25],'XLim',[0 1.4]);
147 legend({'\mu_{max}','K-SVD','Grassmannian','INK-SVD','\mu_{min}'});
148 xlabel('\mu');
149 ylabel('SNR (dB)');
150
151 figure
152 %subplot(3,1,3)
153 hold on, grid on
154 title('Gabor Initialisation')
155 plot([1 1],[0 25],'k-');
156 plot(mu(25),sigNoiseRatio(25),'ks');
157 plot(mu(36),sigNoiseRatio(36),'kd');
158 plot(mu(26:35),sigNoiseRatio(26:35),'k*-');
159 plot([minMu minMu],[0 25],'k--')
160 set(gca,'YLim',[0 25],'XLim',[0 1.4]);
161 legend({'\mu_{max}','K-SVD','Grassmannian','INK-SVD','\mu_{min}'});
162 xlabel('\mu');
163 ylabel('SNR (dB)');
164
165 % minMu = sqrt((dictSize-blockSize)/(blockSize*(dictSize-1)));
166 % maxSNR = max(sigNoiseRatio);
167 %
168 % figure, subplot(2,2,1)
169 % snrMat = buffer(sigNoiseRatio(1:9),3);
170 % bar(snrMat');
171 % title('Signal to noise ratio')
172 % xlabel('Initial dictionary')
173 % ylabel('SNR (dB)')
174 % set(gca,'XTickLabel',{'data','dct','gabor'});
175 % legend('none','Mailhe','Tropp')
176 % grid on
177 %
181 % subplot(2,2,2), grid on 178 % subplot(2,2,2), grid on
182 % snrMat = buffer(sigNoiseRatio(10:18),3); 179 % snrMat = buffer(sigNoiseRatio(10:18),3);
183 % bar(snrMat'); 180 % bar(snrMat');
184 % title('SNR - Mailhe Update') 181 % title('SNR - Mailhe Update')
185 % xlabel('Initial dictionary') 182 % xlabel('Initial dictionary')
186 % ylabel('SNR (dB)') 183 % ylabel('SNR (dB)')
187 % set(gca,'XTickLabel',{'data','dct','gabor'},'YLim',[0 maxSNR+1]); 184 % set(gca,'XTickLabel',{'data','dct','gabor'},'YLim',[0 maxSNR+1]);
188 % legend('none','Mailhe','Tropp') 185 % legend('none','Mailhe','Tropp')
189 % grid on 186 % grid on
190 187 %
191 subplot(1,2,2), hold on, grid on 188 % subplot(2,2,3), hold on, grid on
192 title('Coherence') 189 % title('Coherence')
193 muMat = buffer(mu(1:9),3); 190 % muMat = buffer(mu(1:9),3);
194 line([0.5 3.5],[1 1],'Color','r'); 191 % line([0.5 3.5],[1 1],'Color','r');
195 bar(muMat'); 192 % bar(muMat');
196 line([0.5 3.5],[minMu minMu],'Color','k'); 193 % line([0.5 3.5],[minMu minMu],'Color','k');
197 set(gca,'XTick',1:3,'XTickLabel',{'data','dct','gabor'},'YLim',[0 1.05]) 194 % set(gca,'XTick',1:3,'XTickLabel',{'data','dct','gabor'},'YLim',[0 1.05])
198 legend('\mu_{max}','none','Mailhe','Tropp','\mu_{min}') 195 % legend('\mu_{max}','none','Mailhe','Tropp','\mu_{min}')
199 ylabel('\mu') 196 % ylabel('\mu')
200 xlabel('Initial dictionary') 197 % xlabel('Initial dictionary')
201 198 %
202 % subplot(2,2,4), hold on, grid on 199 % subplot(2,2,4), hold on, grid on
203 % title('Coherence - Mailhe Update') 200 % title('Coherence - Mailhe Update')
204 % muMat = buffer(mu(10:18),3); 201 % muMat = buffer(mu(10:18),3);
205 % line([0.5 3.5],[1 1],'Color','r'); 202 % line([0.5 3.5],[1 1],'Color','r');
206 % bar(muMat'); 203 % bar(muMat');