doFirstStep.m

function [trainedModel,trainingModel,normalizeValues,objectImage,bset,oset,bsetaux,deleted] = doFirstStep(I,rectangleContenidor,numberBlocksY,numberBlocksX,trainingModel,secondAproximationType)
    [nr,nc]= size(I(:,:,1));
    windowSizeY = floor(nr/numberBlocksY);
    windowSizeX = floor(nc/numberBlocksX);
    stuffY = windowSizeY - mod(nr,windowSizeY);
    stuffX = windowSizeX - mod(nc,windowSizeX);
    I = padarray(I, [stuffY stuffX], 'replicate','post');
    [nr,nc]= size(I(:,:,1));

    numberBlocks = nr/windowSizeY * nc/windowSizeX;
    trainingDataset = zeros([numberBlocks 17]);
    charBackground = 'B';
    charObject = 'O';
    prediction = charBackground;

    %HISTOGRAMA
    bset = zeros([1 17]);

    cont = 0;
    for i=1:windowSizeY:nr
        for j=1:windowSizeX:nc

            cont = cont + 1;
            B = I(i:i+windowSizeY-1,j:j+windowSizeX-1,:);

            featureVector=computeFeatures(B);
            trainingDataset(cont,:)=featureVector;

            if blockInsideRectangle([j i windowSizeX windowSizeY],rectangleContenidor,0.8)
                prediction = [prediction; charObject];
            else
                bset = [bset;featureVector];
                prediction = [prediction; charBackground];
            end
        end
    end
    prediction = prediction(2:(numberBlocks+1));


    [normalizeValues,normalizedTrainingDataset,deleted] = normalizeColumns(trainingDataset);
    %Train model to differentiate between the two type of blocks
    trainedModel = trainModel(normalizedTrainingDataset,prediction,trainingModel);

    %Predict model for the blocks inside the rectangle
    [oset,bsetaux,predictedBlocks] = predictModel(I,trainedModel,rectangleContenidor,windowSizeX,windowSizeY,trainingModel,normalizeValues,deleted);

    firstCleaning = predictedBlocks == 1;
    firstCleaning = deleteInteriorHoles(firstCleaning);
    firstCleaning = deleteLittleRegions(firstCleaning,windowSizeY,windowSizeX);

    [expandedEdges,interior] = expandEdges(firstCleaning,windowSizeY,windowSizeX);

    predictedPoints = predictPoints(I,expandedEdges,trainedModel,trainingModel,normalizeValues,secondAproximationType,deleted);

    aux = or(predictedPoints,interior);
    aux = deleteInteriorHoles(aux);
    figure;imshow(aux,[]);

    [finalImage, objectImage] = doFinalImage(I,aux);
    figure;imshow(finalImage,[]);
end