| function featurevectors=constructhf(inputvectors, map) | |
| %construct rotation invariant features from uniform LBP histogram | |
| %inputvectors: NxD array, N histograms of D bins each | |
| %map: mapping struct from getmaphf | |
| % | |
| %EXAMPLE: | |
| %I=imread('rice.png'); | |
| %I2=imrotate(I,90); | |
| %mapping=getmaplbphf(8); | |
| %h=lbp(I,1,8,mapping,'h'); | |
| %h=h/sum(h); | |
| %histograms(1,:)=h; | |
| %h=lbp(I2,1,8,mapping,'h'); | |
| %h=h/sum(h); | |
| %histograms(2,:)=h; | |
| %lbp_hf_features=constructhf(histograms,mapping); | |
| % | |
| %The two rows of lbp_hf_features now contain LBP | |
| %histogram Fourier feature vectors of rice.png and | |
| %its rotated version (with LBP radius 1 and 8 sampling | |
| %points) | |
| n=map.samples; | |
| FVLEN=(n-1)*(floor(n/2)+1)+3; | |
| featurevectors=zeros(size(inputvectors,1),FVLEN); | |
| k=1; | |
| for j=1:length(map.orbits) | |
| b=inputvectors(:,map.orbits{j}+1); | |
| if(size(b,2) > 1) | |
| b=fft(b')'; | |
| b=abs(b); | |
| b=b(:,1:(floor(size(b,2)/2)+1)); | |
| end | |
| featurevectors(:,k:k+size(b,2)-1)=b; | |
| k=k+size(b,2); | |
| end | |