Training SVM using HOG features using JAVA

asked 2017-02-15 18:31:31 -0600

Tomna gravatar image

updated 2020-11-05 15:04:22 -0600

i want to pass a license plate to the SVM and classify either a plate or not

but i still didnt get enough images to train my SVM so i am using another database which is cars from such link ( which consists of 1050 images i divided the positive(550) and negative(500) on different files

  protected static final String PATH_POSITIVE = "/home/tomna/NetBeansProjects/main/src/main/PostiveTrain";
    protected static final String PATH_NEGATIVE = "/home/tomna/NetBeansProjects/main/src/main/NegativeTrain";
    protected static final String XML = "/home/tomna/NetBeansProjects/main/src/main/data.xml";
    protected static final String FILE_TEST = "/home/tomna/NetBeansProjects/main/src/main/Dadcar3.png";

    private static Mat trainingImages = new Mat();
    private static Mat trainingLabels = new Mat();
    private static Mat trainingData = new Mat();
    private static Mat classes = new Mat();
    private static SVM clasificador= SVM.create();

    public static MatOfPoint loacation = new MatOfPoint();
    public static MatOfDouble descriptors = new MatOfDouble();
    private static HOGDescriptor hog;

here i am looping on the folder of positive images to identify such samples are positive

public static void trainPositives() {

    for (File file : new File(PATH_POSITIVE).listFiles()) {
        Mat img = getMat(file.getAbsolutePath());
        trainingImages.push_back(img.reshape(1, 1));
        trainingLabels.push_back(Mat.ones(new Size(1, 1), CvType.CV_32FC1));

as well as for the negative images

public static void trainNegatives() {
    for (File file : new File(PATH_NEGATIVE).listFiles()) {
        Mat img = getMat(file.getAbsolutePath());
        trainingImages.push_back(img.reshape(1, 1));
        trainingLabels.push_back(Mat.zeros(new Size(1, 1), CvType.CV_32FC1));


My question when it comes to such part how am i supposed to get HOG features from each and every image and save it to be used for the SVM training

i tried such code on a single image as in to extract the images features but i didnt understand the output of the image MAT Mat [ 01CV_64FC1, isCont=true, isSubmat=false, nativeObj=0x7f332461cd90, dataAddr=0x0 ]

  public static MatOfPoint loacation = new MatOfPoint();
    private static HOGDescriptor hog= new HOGDescriptor();
     public static MatOfDouble descriptors = new MatOfDouble();
    void run() throws IOException {
        Mat ro = Imgcodecs.imread("/home/tomna/NetBeansProjects/main/src/main/PostiveTrain/pos-1.pgm");
        Imgproc.resize(ro, ro, new Size(64,128));
    hog.detect(ro, loacation, descriptors);

i didn't yet train or test my SVM because i think the features should be extracted properly first but here is an attempt not complete

    public static void test() {
        Mat input = Imgcodecs.imread(new File(FILE_TEST).getAbsolutePath()); File(XML).getAbsolutePath());
        Mat output = new Mat();
        input.convertTo(output, CV_32FC1);
        output = output.reshape(64, 64);


    public static void train() {

        trainingData.convertTo(trainingData, CV_32FC1);
//        hog.setSVMDetector(classes);
//        hog.compute(classes, new Size(16,16),);
//        clasificador.train(trainingData, 0, classes);
edit retag flag offensive close merge delete


There is a sample explaining in nice detail how HOG has to be calculated as feature. Take a look here:

StevenPuttemans gravatar imageStevenPuttemans ( 2017-02-16 08:27:32 -0600 )edit