% author: M.M.Haji
% email : mehdi.haji@gmail.com
%
% generate training data for the text concept.
% if the window contains text, then target is 1
% otherwise the target is 0.

clear

kernel_w = 8; % width of kernel (window)
kernel_h = 8; % height of kernel (window)

dct18_coefs = [4 5 6 12 13 14 20 21 22 44 45 46 52 53 54 60 61 62];

fid = fopen('textdata.txt','w');

tr_images = { 'newcnn512x512.jpg', 'image048.jpg', 'shapes.jpg', 'npaper01.JPG', 'a01-003.jpg'};
%       'image031.jpg'
%       'text-graphics.JPG'

% write [0 0 0 0 ... 0] as the first training data:
for i4 = 1:18
    fprintf(fid,'%f ',0);
end
fprintf(fid,'%d\r\n',0);
   
for i3 = 1:length(tr_images)
    
    fname = tr_images{i3};
    disp(fname);
    I = imread(fname);
    M = imread(strcat(fname(1:find(fname=='.')-1),'_mask.bmp'));
    
    if ndims(I) == 3
        I = rgb2gray(I);
    end
    
    [nr nc] = size(I);

%     NR = ceil(nr/kernel_h) * kernel_h;
%     NC = ceil(nc/kernel_w) * kernel_h;
%     I = padarray(I,[NR-nr NC-nc],0,'post');
%     M = padarray(M,[NR-nr NC-nc],0,'post');
  
    for i1 = 1:kernel_h/2:nr-kernel_h
        for i2 = 1:kernel_w/2:nc-kernel_w
            J = I(i1:i1+kernel_h-1, i2:i2+kernel_w-1);
            % imshow(J);
            
            J = dct2(J);
            J = J';
            J = J(:);
            
            if any(J(dct18_coefs(1:18)))
                for i4 = 1:18
                    fprintf(fid,'%f ',J(dct18_coefs(i4)));
                end
                
                if sum(sum(M(i1:i1+kernel_h-1, i2:i2+kernel_w-1))) >= (kernel_h*kernel_w/2)
                    fprintf(fid,'%d\r\n',1);
                else
                    fprintf(fid,'%d\r\n',0);
                end
            end

        end
    end
end

fclose(fid);
% [str, f1, f2, f3, f4, f5, f6, f7, f8, f9, f10, f11, f12, f13, f14, f15, f16, f17, f18, target] = textread('textdata.txt','%s %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %f, %d;');