Mercurial > hg > smallbox

comparison examples/SMALL_test_coherence.m @ 156:a4d0977d4595 danieleb

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
First branch commit, danieleb
author danieleb
date Tue, 30 Aug 2011 11:12:31 +0100
parents
children 8324c7ea6602
comparison
equal deleted inserted replaced
153:af307f247ac7 156:a4d0977d4595
1 clear
2
3 %% Parameteres
4
5 % Dictionary learning parameters
6 toolbox = 'TwoStepDL'; %dictionary learning toolbox
7 dicUpdate = {'ksvd','mailhe'}; %dictionary updates
8 iterNum = 20; %number of iterations
9
10 % Test signal parameters
11 signal = audio('music03_16kHz.wav'); %audio signal
12 blockSize = 256; %size of audio frames
13 dictSize = 512; %number of atoms in the dictionary
14 overlap = 0.5; %overlap between consecutive frames
15 sigma = 1e6; %snr of noise (set to be negligible so that the problem becomes approximation rather than denoising)
16 percActiveAtoms = 5; %percentage of active atoms
17
18 % Dependent parameters
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)
21 minCoherence = sqrt((dictSize-blockSize)/(blockSize*(dictSize-1))); %target coherence (based on coherence lower bound)
22
23 % Initial dictionaries
24 dctDict = dictionary('dct',blockSize,dictSize);
25 dctDict = dctDict.phi;
26 gaborParam = struct('N',blockSize,'redundancyFactor',2,'wd',@rectwin);
27 gaborDict = Gabor_Dictionary(gaborParam);
28
29 %% Generate audio denoising problem with low noise (audio representation)
30 SMALL.Problem = generateAudioDenoiseProblem(signal.s,[],blockSize,...
31 dictSize,overlap,sigma); % generate representation problem
32 SMALL.Problem.b1 = SMALL.Problem.b; % copy signals from training set b to test set b1 (needed for later functions)
33
34 % omp2 sparse representation solver
35 ompParam = struct('X',SMALL.Problem.b,'epsilon',epsilon,'maxatoms',nActiveAtoms); %parameters
36 solver = SMALL_init_solver('ompbox','omp2',ompParam,false); %solver structure
37
38
39 %% Test ksvd dictionary update
40 name = dicUpdate{1}; %use ksvd update
41 SMALL.DL(1:9) = SMALL_init_DL(toolbox,name); %create dictionary learning structures
42
43 % learn with random initialisation and no decorrelation
44 SMALL.DL(1).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
45 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
46 'decFcn','none'); %parameters for the dictionary learning
47 SMALL.DL(1) = SMALL_learn(SMALL.Problem,SMALL.DL(1)); %learn dictionary
48
49 % learn with random initialisation and mailhe decorrelation
50 SMALL.DL(2).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
51 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
52 'decFcn','mailhe','coherence',minCoherence); %parameters for the dictionary learning
53 SMALL.DL(2) = SMALL_learn(SMALL.Problem,SMALL.DL(2)); %learn dictionary
54
55 % learn with random initialisation and tropp decorrelation
56 SMALL.DL(3).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
57 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
58 'decFcn','tropp','coherence',minCoherence); %parameters for the dictionary learning
59 SMALL.DL(3) = SMALL_learn(SMALL.Problem,SMALL.DL(3)); %learn dictionary
60
61 % Learn with dct initialisation and no decorrelation
62 SMALL.DL(4).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
63 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
64 'decFcn','none','initdict',dctDict); %parameters for the dictionary learning
65 SMALL.DL(4) = SMALL_learn(SMALL.Problem,SMALL.DL(4)); %learn dictionary
66
67 % learn with dct initialisation and mailhe decorrelation
68 SMALL.DL(5).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
69 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
70 'decFcn','mailhe','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning
71 SMALL.DL(5) = SMALL_learn(SMALL.Problem,SMALL.DL(5)); %learn dictionary
72
73 % learn with dct initialisation and tropp decorrelation
74 SMALL.DL(6).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
75 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
76 'decFcn','tropp','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning
77 SMALL.DL(6) = SMALL_learn(SMALL.Problem,SMALL.DL(6)); %learn dictionary
78
79 % Learn with gabor initialisation and no decorrelation
80 SMALL.DL(7).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
81 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
82 'decFcn','none','initdict',gaborDict); %parameters for the dictionary learning
83 SMALL.DL(7) = SMALL_learn(SMALL.Problem,SMALL.DL(7)); %learn dictionary
84
85 % learn with gabor initialisation and mailhe decorrelation
86 SMALL.DL(8).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
87 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
88 'decFcn','mailhe','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning
89 SMALL.DL(8) = SMALL_learn(SMALL.Problem,SMALL.DL(8)); %learn dictionary
90
91 % learn with gabor initialisation and tropp decorrelation
92 SMALL.DL(9).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
93 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
94 'decFcn','tropp','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning
95 SMALL.DL(9) = SMALL_learn(SMALL.Problem,SMALL.DL(9)); %learn dictionary
96
97 %% Test mailhe dictionary update
98 name = dicUpdate{2}; %use mailhe update
99 SMALL.DL(10:18) = SMALL_init_DL(toolbox,name); %create dictionary learning structure
100
101 % learn with random initialisation and no decorrelation
102 SMALL.DL(10).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
103 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
104 'decFcn','none'); %parameters for the dictionary learning
105 SMALL.DL(10) = SMALL_learn(SMALL.Problem,SMALL.DL(10)); %learn dictionary
106
107 % learn with random initialisation and mailhe decorrelation
108 SMALL.DL(11).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
109 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
110 'decFcn','mailhe','coherence',minCoherence); %parameters for the dictionary learning
111 SMALL.DL(11) = SMALL_learn(SMALL.Problem,SMALL.DL(11)); %learn dictionary
112
113 % learn with random initialisation and tropp decorrelation
114 SMALL.DL(12).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
115 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
116 'decFcn','tropp','coherence',minCoherence); %parameters for the dictionary learning
117 SMALL.DL(12) = SMALL_learn(SMALL.Problem,SMALL.DL(12)); %learn dictionary
118
119 % Learn with dct initialisation and no decorrelation
120 SMALL.DL(13).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
121 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
122 'decFcn','none','initdict',dctDict); %parameters for the dictionary learning
123 SMALL.DL(13) = SMALL_learn(SMALL.Problem,SMALL.DL(13)); %learn dictionary
124
125 % learn with dct initialisation and mailhe decorrelation
126 SMALL.DL(14).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
127 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
128 'decFcn','mailhe','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning
129 SMALL.DL(14) = SMALL_learn(SMALL.Problem,SMALL.DL(14)); %learn dictionary
130
131 % learn with dct initialisation and tropp decorrelation
132 SMALL.DL(15).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
133 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
134 'decFcn','tropp','coherence',minCoherence,'initdict',dctDict); %parameters for the dictionary learning
135 SMALL.DL(15) = SMALL_learn(SMALL.Problem,SMALL.DL(15)); %learn dictionary
136
137 % Learn with gabor initialisation and no decorrelation
138 SMALL.DL(16).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
139 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
140 'decFcn','none','initdict',gaborDict); %parameters for the dictionary learning
141 SMALL.DL(16) = SMALL_learn(SMALL.Problem,SMALL.DL(16)); %learn dictionary
142
143 % learn with gabor initialisation and mailhe decorrelation
144 SMALL.DL(17).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
145 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
146 'decFcn','mailhe','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning
147 SMALL.DL(17) = SMALL_learn(SMALL.Problem,SMALL.DL(17)); %learn dictionary
148
149 % learn with gabor initialisation and tropp decorrelation
150 SMALL.DL(18).param = struct('data',SMALL.Problem.b,'Tdata',nActiveAtoms,...
151 'dictsize',dictSize,'iternum',iterNum,'memusage','high','solver',solver,...
152 'decFcn','tropp','coherence',minCoherence,'initdict',gaborDict); %parameters for the dictionary learning
153 SMALL.DL(18) = SMALL_learn(SMALL.Problem,SMALL.DL(18)); %learn dictionary
154
155 %% Evaluate coherence and snr of representation for the various methods
156 sigNoiseRatio = zeros(18,1);
157 mu = zeros(18,1);
158 for i=1:18
159 SMALL.Problem.A = SMALL.DL(i).D;
160 tempSolver = SMALL_solve(SMALL.Problem,solver);
161 sigNoiseRatio(i) = snr(SMALL.Problem.b,SMALL.DL(i).D*tempSolver.solution);
162 dic(i) = dictionary(SMALL.DL(i).D);
163 mu(i) = dic(i).coherence;
164 end
165
166 %% Plot results
167 minMu = sqrt((dictSize-blockSize)/(blockSize*(dictSize-1)));
168 maxSNR = max(sigNoiseRatio);
169
170 figure, subplot(2,2,1)
171 snrMat = buffer(sigNoiseRatio(1:9),3);
172 bar(snrMat');
173 title('SNR - KSVD Update')
174 xlabel('Initial dictionary')
175 ylabel('SNR (dB)')
176 set(gca,'XTickLabel',{'data','dct','gabor'},'YLim',[0 maxSNR+1]);
177 legend('none','Mailhe','Tropp')
178 grid on
179
180 subplot(2,2,2), grid on
181 snrMat = buffer(sigNoiseRatio(10:18),3);
182 bar(snrMat');
183 title('SNR - Mailhe Update')
184 xlabel('Initial dictionary')
185 ylabel('SNR (dB)')
186 set(gca,'XTickLabel',{'data','dct','gabor'},'YLim',[0 maxSNR+1]);
187 legend('none','Mailhe','Tropp')
188 grid on
189
190 subplot(2,2,3), hold on, grid on
191 title('Coherence - KSVD Update')
192 muMat = buffer(mu(1:9),3);
193 line([0.5 3.5],[1 1],'Color','r');
194 bar(muMat');
195 line([0.5 3.5],[minMu minMu],'Color','k');
196 set(gca,'XTick',1:3,'XTickLabel',{'data','dct','gabor'},'YLim',[0 1.05])
197 legend('\mu_{max}','none','Mailhe','Tropp','\mu_{min}')
198 ylabel('\mu')
199 xlabel('Initial dictionary')
200
201 subplot(2,2,4), hold on, grid on
202 title('Coherence - Mailhe Update')
203 muMat = buffer(mu(10:18),3);
204 line([0.5 3.5],[1 1],'Color','r');
205 bar(muMat');
206 line([0.5 3.5],[minMu minMu],'Color','k');
207 set(gca,'XTick',1:3,'XTickLabel',{'data','dct','gabor'},'YLim',[0 1.05])
208 legend('\mu_{max}','none','Mailhe','Tropp','\mu_{min}')
209 ylabel('\mu')
210 xlabel('Initial dictionary')