function SVMRFE_inx = svmrfe(trn_data,trn_cls)

% Input:
%     trn_data: sample_num x feat_num matrix
%     trn_cls: sample_num x 1 matrix
% Output:
%     SVMRFE_inx: output index ranked by SVMRFE
% Note: 
%     This code calls RfeRanking.m, which is provided by Dr.MATTEO
%     MASOTTI at http://www.bo.infn.it/~masotti/MySoftware/RFE/RfeRanking.m
%
% This code was originally designed for the feature selection in image 
% steganalysis with binary classification of the following paper:
%
% Q. Liu, A. Sung, Z. Chen, J. Xu (2008). Feature mining and pattern
% classification for steganalysis of LSB matching steganography in
% grayscale images. Pattern Recognition  41 (1): 56-66. 
%
% 
% The function is re-written by Qingzhong Liu @ 06/04/2009. Please cite the
% above paper when you use this code.
% 
%

kertype = 0; %% 0 --- Linear. You may adjust the parameters in the following lines.
Parameters(1) = kertype;
Parameters(2) = 1;
Parameters(3) = 1;
Parameters(4) = 0;
Parameters(5) = 1;
Parameters(6) = 10;
Parameters(7) = 0.001;    

[smaple_num, feat_dim] = size(trn_data);

if smaple_num ~= length(trn_cls)
    disp('No matching between sample number and class number, ERROR!');
    disp('Please correct and try again!');
    return 
end

%%%%%%%%%%%%
cls_num = unique(trn_cls);

if length(cls_num) > 2
    disp('More than two class numbers, ERROR!');
    disp('Please correct and try again!');
    return 
end
   
inx1 = find(trn_cls == cls_num(1));
inx2 = find(trn_cls == cls_num(2));
label_trn = trn_cls; 

label_trn(inx1) = -1;   
label_trn(inx2) = 1;     


[AlphaY, SVs, Bias, Parameters, nSV, nLabel] = SVMTrain(trn_data', label_trn', Parameters); % SVM training, call OSU-SVM function
[rank_feat, ind_feat] = RfeRanking(AlphaY, SVs, Parameters);         % call RfeRanking

ind_feat2(1:feat_dim,1) = [1:feat_dim]';
ind_feat2(1:feat_dim,2) = ind_feat';

for k = feat_dim-1:-1:2
    orig_inx = ind_feat(1:k);
    RFE_data = trn_data(:,orig_inx);
    
    [AlphaY, SVs, Bias, Parameters, nSV, nLabel] = SVMTrain(RFE_data', label_trn', Parameters); % Recursively call SVM training
    [rank_feat2, ind2] = RfeRanking(AlphaY, SVs, Parameters); % Recursively call RfeRanking
    for m = 1:length(ind2)
        inx = ind2(m);
        ind_feat(m) = ind_feat2(inx,2);
    end
    
    ind_feat2(1:feat_dim,1) = [1:feat_dim]';
    ind_feat2(1:feat_dim,2) = ind_feat';
    fprintf('SVMRFE feature selection, the %8d -th element feature subset determined!\n',k);
end

SVMRFE_inx = ind_feat;

return