Demo illustrating Circular Histogram Local Binary Pattern (chlbp)
Circular Histogram Local Binary Pattern.
Usage
------
H = chlbp(I , [N] , [R] , [map] , [shiftbox] );
Inputs
-------
I Input image (ny x nx x P) in UINT8 format.
N Number of sampling points (1 x nR) (default [N=8])
R Vector of Radius (1 x nR) (default [R=1]) ny>2max(R)+1 & nx>2max(R)+1
map Mapping of the chlbp (2^Nmax x nR) in double format (default map = (0 : 255))
shiftbox Shifting box parameters shiftbox (2 x 2 x nR) where [by , bx ; deltay , deltax] x nR (default shiftbox = [ny , nx ; 0 , 0])
by, bx denote the size of subwindows analysis, deltay, deltax represent the shift the subwindows analysis. Outputs
-------
H chlbp features (bin*nH x P) int UINT32 format where nH = nR*max(floor((ny - shiftbox(1,1,:))./shiftbox(2,1,:)) + 1 , 1)*max(floor((nx - shiftbox(1,2,:))./shiftbox(2,2,:)) + 1 , 1)
Contents
First example : compute Uniform CHLBP Features with N =8 neighbours and Radius R = 1 with a subwindows analysis of 64x64
shifted by 16x16
clc,clear, close all,drawnow
I = (imread('rice.png'));
N = 8;
R = 1;
mapping = getmapping(N,'u2');
map = mapping.table';
shiftbox = [64 , 64 ; 16 , 16];
H = chlbp(I , N , R , map , shiftbox);
plot(H)
title(sprintf('Uniform CHLBP Histogram with N = %d, R = %d' , N , R))
disp('Press key to continue')
pause
Warning: Calling MEX-file 'C:\utilisateurs\SeBy\matlab\fdtool\chlbp.dll'.
MEX-files with .dll extensions will not execute in a future version of MATLAB.
Press key to continue
Second example : compute CHLBP_{8;1}^u Features on Viola-Jones database
load viola_24x24
Ny = 24;
Nx = 24;
N = 8;
R = 1;
mapping = getmapping(N,'u2');
map = mapping.table';
H = chlbp(X , N , R , map);
imagesc(H)
title(sprintf('Uniform CHLBP Histogram with N = %d, R = %d' , N , R))
disp('Press key to continue')
pause
Press key to continue
Third example : compute CHLBP_{8;1}^u + CHLBP_{4;1} Features
Ny = 19;
Nx = 19;
X = uint8(ceil(256*rand(Ny , Nx)));
N = [8 , 4];
R = [1 , 1];
map = zeros(2^max(N) , length(N));
mapping = getmapping(N(1),'u2');
map(: , 1) = mapping.table';
map(1:2^N(2) , 2) = (0:2^N(2)-1)';
shiftbox = cat(3 , [Ny , Nx ; 0 , 0] , [10 , 10 ; 4 , 4]);
H = chlbp(X , N , R , map , shiftbox);
plot(H)
title('CHLBP_{8;1}^u + CHLBP_{4;1} Histogram')
disp('Press key to continue')
pause
Warning: Out of range value converted to intmin('uint8') or intmax('uint8').
Press key to continue
Fourth example : compute CHLBP_{8;1}^u + CHLBP_{4;1} Features + Adaboosting with T weaklearners (Decision Stump)
load viola_24x24
Ny = 24;
Nx = 24;
N = [8 , 4];
R = [1 , 1];
map = zeros(2^max(N) , length(N));
mapping = getmapping(N(1),'u2');
map(1:2^N(1) , 1) = mapping.table';
map(1:2^N(2) , 2) = (0:2^N(2)-1)';
shiftbox = cat(3 , [Ny , Nx ; 1 , 1] , [16 , 16 ; 4 , 4]);
T = 50;
H = chlbp(X , N , R , map , shiftbox);
figure
imagesc(H)
title('CHLBP Features')
drawnow
y = int8(y);
indp = find(y == 1);
indn = find(y ==-1);
index = randperm(length(y));
model = chlbp_adaboost_binary_model_cascade(H(: , index) , y(index) , T);
[yest_train , fx_train] = chlbp_adaboost_binary_predict_cascade(H , model);
tp_train = sum(yest_train(indp) == y(indp))/length(indp)
fp_train = 1 - sum(yest_train(indn) == y(indn))/length(indn)
Perf_train = sum(yest_train == y)/length(y)
[dum , ind] = sort(y , 'descend');
figure
plot(fx_train(ind))
title(sprintf('Output of the strong classifier for train data with T = %d' , T))
[tpp_train , fpp_train] = basicroc(y , fx_train);
load jensen_24x24
y = int8(y);
indp = find(y == 1);
indn = find(y ==-1);
H = chlbp(X , N , R , map , shiftbox);
[yest_test , fx_test] = chlbp_adaboost_binary_predict_cascade(H , model);
tp_test = sum(yest_test(indp) == y(indp))/length(indp)
fp_test = 1 - sum(yest_test(indn) == y(indn))/length(indn)
Perf_test = sum(yest_test == y)/length(y)
[dum , ind] = sort(y , 'descend');
figure
plot(fx_test(ind))
title(sprintf('Output of the strong classifier for test data with T = %d' , T))
[tpp_test , fpp_test] = basicroc(y , fx_test);
figure
plot(fpp_train , tpp_train , fpp_test , tpp_test , 'r' , 'linewidth' , 2)
axis([-0.02 , 1.02 , -0.02 , 1.02])
title(sprintf('ROC for CHLBP features with T = %d' , T))
legend('Train' , 'Test' , 'Location' , 'SouthEast')
disp('Press key to continue')
pause
Warning: Calling MEX-file 'C:\utilisateurs\SeBy\matlab\fdtool\chlbp_adaboost_binary_model_cascade.dll'.
MEX-files with .dll extensions will not execute in a future version of MATLAB.
Warning: Calling MEX-file 'C:\utilisateurs\SeBy\matlab\fdtool\chlbp_adaboost_binary_predict_cascade.dll'.
MEX-files with .dll extensions will not execute in a future version of MATLAB.
tp_train =
0.9463
fp_train =
0.0351
Perf_train =
0.9578
tp_test =
0.9642
fp_test =
0.0340
Perf_test =
0.9657
Press key to continue
Fifth example : compute CHLBP_{8;1}^u + CHLBP_{4;1} + CHLBP_{12;2}^u Features + Gentleboosting with T weaklearners (Decision
Stump)
load viola_24x24
Ny = 24;
Nx = 24;
N = [8 , 4 , 12];
R = [1 , 1 , 2];
map = zeros(2^max(N) , length(N));
mapping = getmapping(N(1),'u2');
map(1:2^N(1) , 1) = mapping.table';
map(1:2^N(2) , 2) = (0:2^N(2)-1)';
mapping = getmapping(N(3),'u2');
map(1:2^N(3) , 3) = mapping.table';
shiftbox = cat(3 , [Ny , Nx ; 1 , 1] , [16 , 16 ; 4 , 4] , [Ny , Nx ; 1 , 1]);
T = 50;
H = chlbp(X , N , R , map , shiftbox);
figure
imagesc(H)
title('CHLBP Features')
drawnow
y = int8(y);
indp = find(y == 1);
indn = find(y ==-1);
index = randperm(length(y));
model = chlbp_gentleboost_binary_model_cascade(H(: , index) , y(index) , T);
[yest_train , fx_train] = chlbp_gentleboost_binary_predict_cascade(H , model);
tp_train = sum(yest_train(indp) == y(indp))/length(indp)
fp_train = 1 - sum(yest_train(indn) == y(indn))/length(indn)
Perf_train = sum(yest_train == y)/length(y)
[dum , ind] = sort(y , 'descend');
figure
plot(fx_train(ind))
title(sprintf('Output of the strong classifier for train data with T = %d' , T))
[tpp_train , fpp_train] = basicroc(y , fx_train);
load jensen_24x24
y = int8(y);
indp = find(y == 1);
indn = find(y ==-1);
H = chlbp(X , N , R , map , shiftbox);
[yest_test , fx_test] = chlbp_gentleboost_binary_predict_cascade(H , model);
tp_test = sum(yest_test(indp) == y(indp))/length(indp)
fp_test = 1 - sum(yest_test(indn) == y(indn))/length(indn)
Perf_test = sum(yest_test == y)/length(y)
[dum , ind] = sort(y , 'descend');
figure
plot(fx_test(ind))
title(sprintf('Output of the strong classifier for test data with T = %d' , T))
[tpp_test , fpp_test] = basicroc(y , fx_test);
figure
plot(fpp_train , tpp_train , fpp_test , tpp_test , 'r' , 'linewidth' , 2)
axis([-0.02 , 1.02 , -0.02 , 1.02])
title(sprintf('ROC for CHLBP features with T = %d' , T))
legend('Train' , 'Test', 'Location' , 'SouthEast')
disp('Press key to continue')
pause
Warning: Calling MEX-file 'C:\utilisateurs\SeBy\matlab\fdtool\chlbp_gentleboost_binary_model_cascade.dll'.
MEX-files with .dll extensions will not execute in a future version of MATLAB.
Warning: Calling MEX-file 'C:\utilisateurs\SeBy\matlab\fdtool\chlbp_gentleboost_binary_predict_cascade.dll'.
MEX-files with .dll extensions will not execute in a future version of MATLAB.
tp_train =
0.9585
fp_train =
0.0227
Perf_train =
0.9701
tp_test =
0.9804
fp_test =
0.0244
Perf_test =
0.9765
Press key to continue
Sith example : CHLBP_{8;1}^u versus CHLBP_{8;1} Features + Gentleboosting with T weaklearners (Decision Stump)
load viola_24x24
y = int8(y);
indp = find(y == 1);
indn = find(y ==-1);
Ny = 24;
Nx = 24;
N = 8;
R = 1;
map = zeros(2^max(N) , length(N));
mapping = getmapping(N(1),'u2');
map0 = mapping.table';
map1 = (0:2^N-1)';
T = 50;
H0 = chlbp(X , N , R , map0);
H1 = chlbp(X , N , R , map1);
figure
imagesc(H0)
title(sprintf('Uniform CHLBP Features with N = %d, R = %2.1f' , N , R))
drawnow
figure
imagesc(H1)
title(sprintf('CHLBP Features with N = %d, R = %2.1f' , N , R))
drawnow
index = randperm(length(y));
model0 = chlbp_gentleboost_binary_model_cascade(H0(: , index) , y(index) , T);
[yest0_train , fx0_train] = chlbp_gentleboost_binary_predict_cascade(H0 , model0);
model1 = chlbp_gentleboost_binary_model_cascade(H1(: , index) , y(index) , T);
[yest1_train , fx1_train] = chlbp_gentleboost_binary_predict_cascade(H1 , model1);
tp0_train = sum(yest0_train(indp) == y(indp))/length(indp)
fp0_train = 1 - sum(yest0_train(indn) == y(indn))/length(indn)
Perf0_train = sum(yest0_train == y)/length(y)
tp1_train = sum(yest1_train(indp) == y(indp))/length(indp)
fp1_train = 1 - sum(yest1_train(indn) == y(indn))/length(indn)
Perf1_train = sum(yest1_train == y)/length(y)
[dum , ind] = sort(y , 'descend');
figure
plot((1:length(y)) , fx0_train(ind) , (1:length(y)) , fx1_train(ind) , 'r')
title(sprintf('Output of the strong classifier for train data with T = %d' , T))
legend('Uniform CHLBP' , 'CHLBP')
[tpp0_train , fpp0_train] = basicroc(y , fx0_train);
[tpp1_train , fpp1_train] = basicroc(y , fx1_train);
load jensen_24x24
y = int8(y);
indp = find(y == 1);
indn = find(y ==-1);
H0 = chlbp(X , N , R , map0);
H1 = chlbp(X , N , R , map1);
[yest0_test , fx0_test] = chlbp_gentleboost_binary_predict_cascade(H0 , model0);
[yest1_test , fx1_test] = chlbp_gentleboost_binary_predict_cascade(H1 , model1);
tp0_test = sum(yest0_test(indp) == y(indp))/length(indp)
fp0_test = 1 - sum(yest0_test(indn) == y(indn))/length(indn)
Perf0_test = sum(yest0_test == y)/length(y)
tp1_test = sum(yest1_test(indp) == y(indp))/length(indp)
fp1_test = 1 - sum(yest1_test(indn) == y(indn))/length(indn)
Perf1_test = sum(yest1_test == y)/length(y)
[dum , ind] = sort(y , 'descend');
figure
plot((1:length(y)) , fx0_test(ind) , (1:length(y)) , fx1_test(ind) , 'r')
title(sprintf('Output of the strong classifier for test data with T = %d' , T))
legend('Uniform CHLBP' , 'CHLBP')
[tpp0_test , fpp0_test] = basicroc(y , fx0_test);
[tpp1_test , fpp1_test] = basicroc(y , fx1_test);
figure
plot(fpp0_train , tpp0_train , 'b' , fpp0_test , tpp0_test , 'r' , fpp1_train , tpp1_train , 'b--' , fpp1_test , tpp1_test , 'r--' , 'linewidth' , 2)
axis([-0.02 , 1.02 , -0.02 , 1.02])
title(sprintf('ROC for U-CHLBP and CHLBP features with T = %d' , T))
legend('Train U-CHLBP' , 'Test U-CHLBP' , 'Train CHLBP' , 'Test CHLBP' , 'Location' , 'SouthEast')
tp0_train =
0.9203
fp0_train =
0.0591
Perf0_train =
0.9330
tp1_train =
0.9146
fp1_train =
0.0535
Perf1_train =
0.9342
tp0_test =
0.9430
fp0_test =
0.0428
Perf0_test =
0.9547
tp1_test =
0.9446
fp1_test =
0.0515
Perf1_test =
0.9478
