post_classification.m

function [ out,Probability ] = post_classification( img,Tr,Te )
[no_lines, no_rows, no_bands] = size(img);
%% feature extraction
fimg1=average_fusion(img,40);
fimg2 = ToVector(fimg1);
fimg2 = fimg2';
fimg2=double(fimg2);
%% Training and testing samples
train_SL=matricetotwo(Tr);
train_samples = fimg2(:,train_SL(1,:))';
train_labels= train_SL(2,:)';

test_SL=matricetotwo(Te);
test_samples = fimg2(:,test_SL(1,:))';
test_labels = test_SL(2,:)';
%% Spectral classifier
% Normalizing Training and original img 
[train_samples,M,m] = scale_func(train_samples);
[fimg3 ] = scale_func(fimg2',M,m);
% Selecting the paramter for SVM
[Ccv Gcv cv cv_t]=cross_validation_svm(train_labels,train_samples);
% Training using a Gaussian RBF kernel
%give the parameters of the SVM (Thanks Pedram for providing the
% parameters of the SVM)
parameter=sprintf('-s 0 -t 2 -c %f -g %f -m 500 -b 1',Ccv,Gcv); 
%%% Train the SVM
model=svmtrain(train_labels,train_samples,parameter);
[Result, accuracy,prob] = svmpredict(ones(no_lines*no_rows,1) ,fimg3,model,'-b 1');  
prob=reshape(prob,[no_lines no_rows max(Result(:))]);
% OPimg=PCA_img(img,3);
OPimg =kpca(img, 1000,3, 'Gaussian',20);
% OPimg=SuPCA(img,200,3);
OP_img=reshape(OPimg,[size(OPimg,1)*size(OPimg,2) size(OPimg,3)]);
[ERWresult,Probability] = ERW_OP(OP_img,train_SL,prob);
out=ERWresult';

% Fusion_result=MRF_opt(3,no_lines,no_rows,prob,train_labels',train_SL,max(Result(:)));
end