%for chi-square similarity matrix
A=zeros(26,26)
%for Bhattacharyya similarity matrix
B=zeros(26,26)

for i=1:25
    f=filter(ones(1,50)/50,1,y(i,:))
    %plot de-noise figures
    figure(3),subplot(5,5,i),plot(x(i,:),f)
    switch(i)
        case 1, s='dinosur267'
        case 2, s='dinosur268'
        case 3, s='dinosur269'
        case 4, s='dinosur274'
        case 5, s='dinosur276'
        case 6, s='dog87'
        case 7, s='dog88'
        case 8, s='dog89'
        case 9, s='dog90'
        case 10, s='dog91'
        case 11, s='face293'
        case 12, s='face294'
        case 13, s='face301'
        case 14, s='face313'
        case 15, s='face319'
        case 16, s='horse103'
        case 17, s='horse104'
        case 18, s='horse105'
        case 19, s='horse107'
        case 20, s='horse108'
        case 21, s='plane1122'
        case 22, s='plane1145'
        case 23, s='plane1170'
        case 24, s='plane1180'
        case 25, s='plane1191'
        otherwise
    end
    title(strcat(s,' (',int2str(i),')'))

    %calculate distance between f and g
    for j=i:25
        g=filter(ones(1,50)/50,1,y(j,:))
        %chi-square x^2: D(f,g)=inv((f-g)*(f-g)/(f+g))
        d=(f-g).^2./(f+g)
        %When f[i]==0 and g[i]==0, above equation will has d[i]=NaN because
        %the divider is 0.
        %We need to deal with these NaN before we step further.
        for k=1:1025
            if isnan(d(k))
                d(k)=0
            end
        end            
        A(i,j)=sum(d)
        A(j,i)=A(i,j)
        %Bhattacharyya: D(f,g)=1-inv(sqrt(f*g))
        d=sqrt(f.*g)
        B(i,j)=1-sum(d)
        B(j,i)=B(i,j)
    end
end
%display chi-square distance similarity matrix
figure(4),subplot(1,2,1),colormap gray,pcolor(A),colorbar,title('chi-square statistics distance')
%display Bhattacharyya distance similarity matrix
figure(4),subplot(1,2,2),colormap gray,pcolor(B),colorbar,title('Bhattacharyya distance')