Matlab Face Recognition с использованием SVM выводит тот же класс

Question

В следующем коде реализуется дискриминантный анализ и SVM Фишера с использованием LIBSVM. Я пытаюсь классифицировать тестовое изображение, используя бинарную древовидную структуру и SVM(libsvm), но каждый раз следующий код выводит один и тот же класс. Алгоритм LDA с классификатором EuclideanDistance хорошо acuraccy.

TrainDatabasePath = uigetdir(strcat(matlabroot,'\work'), 'Select training database path' );
TestDatabasePath = uigetdir(strcat(matlabroot,'\work'), 'Select test database path');

%Each column of T contain an training image vector
[T,irow,icol] = CreateDatabase(TrainDatabasePath);
Class_population = 13;
Class_number = ( size(T,2) )/Class_population; 
Train_number=size(T,2);

%LDA IN TRAINING DATABASE
[m,V_PCA,V_Fisher,ProjectedImages_Fisher] =FisherfaceCore(T,Class_population);

prompt = {'Enter test image name (a number between 1 to 50):'};
dlg_title = 'Input of FLD-Based Face Recognition System';
num_lines= 1;
def = {'1'};
TestImage  = inputdlg(prompt,dlg_title,num_lines,def);
TestImage = strcat(TestDatabasePath,'\',char(TestImage),'.jpg');
InputImage = imread(TestImage);

%Project Test Image in Subspace
temp = rgb2gray(InputImage);
temp=double(temp);
[irow icol] = size(temp);
InImage = reshape(temp',irow*icol,1);
Difference = double(InImage)-m; 
ProjectedTestImage = V_Fisher' * V_PCA' * Difference;

%SVM parameters. The kernel is a polynomial
c=1e9;
params=[' -t ' int2str(1) ' -c ' int2str(c)];

% prevArray starts with an array containing each class. Winner array is the
%classes that are selected in the binary tree
prevArr = [1:Class_number]; 
winnerArr = [];
feature=[];
% This section of the code computes a binary SVM tree, by solving a 2 class
% problem with SVM for every 2 classes (1 and 2, 3 and 4, 5 and 6 etc). The
% that you selected is classified according to each 2-class problem and the
% class selected goes on to compete with the other classes selected.
for max = 1:1000
        winnerArr = []; 
        for winRep = 1:2:length(prevArr) 

            % Selects the two classes to train the SVM
            if winRep >= length(prevArr)
                i = prevArr(winRep) ;       
                j = prevArr(winRep-1) ;   
            else
                i = prevArr(winRep)  ;      
                j = prevArr(winRep+1)  ; 
            end

            % Selects the features of the 2 classes
            feature = [ProjectedImages_Fisher(1:2,Class_population*i-(Class_population-1):Class_population*i),ProjectedImages_Fisher(1:2,Class_population*j-(Class_population-1):Class_population*j)]';


            % Assigns the labels for each class
            for m1 = 1:Class_population
               label(m1) = 1; 
            end

            for n1 = Class_population+1:2*Class_population
               label(n1) = -1; 
            end

            % The SVM is trained
            model=svmtrain(label',feature, params);

            % The face that the user selected is classified to any of the two
            % classes
            guessLab(1) = 1;
            predLabel=svmpredict(guessLab',[ProjectedTestImage(1) ProjectedTestImage(2)],model);

            predLabel;

            % A winner class is selected
            if predLabel == 1 
                winnerArr = [winnerArr i];
            elseif predLabel == -1 
                winnerArr = [winnerArr j];
            end

            if winnerArr > 1
                for c1 = 2:length(winnerArr)
                    if winnerArr(c1) == winnerArr(c1-1)
                       winnerArr(c1) = []; 
                    end
                end
            end

        end

        prevArr = winnerArr;

       if length(winnerArr) < 2
           winnerArr;
            break
        end
end

    % This is the class that was selected
    CLASS = winnerArr;
    display('The class that matches your face is:')
    display(CLASS)

https://github.com/JaimeIvanCervantes/FaceRecognition