FaceRecognitionEngine and Local Binary Patterns Histogram

[Q3Varnam]

I am not sure whether there is a ready made option to use LBPH in the FaceRecognitionEngine.

Is there any example available on how to use the FaceRecognitionEngine class?

[Jonathon Hare]

There's currently no proper tutorial, but there is some code you could look at to get an idea here: https://sourceforge.net/p/openimaj/code/HEAD/tree/trunk/tools/FaceTools/src/main/java/org/openimaj/tools/faces/recognition/options/RecognitionStrategy.java (look for the LBPHistograms_KNN enumeration to see how a FaceRecognitionEngine can be pieced together from a face detector, face aligner, LBP feature extractor and KNN classifier.

[Q3Varnam]

Based on your suggestion I did something like this

~~~~~~

static int i = 0;

public static void main(String args[])
{

    VideoCapture video;
    try
    {
        video = new VideoCapture(320, 240);
        video.setFPS(4.0);
        VideoDisplay<MBFImage> display = VideoDisplay
                .createVideoDisplay(video);
        final HaarCascadeDetector fd = new HaarCascadeDetector(
                BuiltInCascade.frontalface_default.classFile(), 120);
        float threshold = Float.NaN;
        int K = 1;
        FloatFVComparison comparison = FloatFVComparison.EUCLIDEAN;
        FacialFeatureComparator<LocalLBPHistogram> comparator = new FaceFVComparator<LocalLBPHistogram, FloatFV>(
                comparison);
        final LocalLBPHistogram.Extractor<DetectedFace> extractor = new LocalLBPHistogram.Extractor<DetectedFace>();
        KNNAnnotator<DetectedFace, String, LocalLBPHistogram> knn = KNNAnnotator
                .create(extractor, comparator, K, threshold);
        AnnotatorFaceRecogniser<DetectedFace, String> recogniser = AnnotatorFaceRecogniser
                .create(knn);
        final FaceRecognitionEngine<DetectedFace, String> engine = FaceRecognitionEngine
                .create(fd, recogniser);

        fd.setMaxSize(300);

        display.addVideoListener(new VideoDisplayListener<MBFImage>() {
            public void beforeUpdate(MBFImage frame)
            {

                List<DetectedFace> faces = fd.detectFaces(Transforms
                        .calculateIntensity(frame));

                for (DetectedFace face : faces)
                {
                    if (i < 10)
                    {
                        engine.train(face, "krishna_" + i);

                        i++;
                    } else
                    {
                        List<IndependentPair<DetectedFace, List<ScoredAnnotation<String>>>> data = engine
                                .recognise(face.getFacePatch());
                        System.out.println(data.size());
                    }
                    frame.drawShape(face.getBounds(), RGBColour.ORANGE);

                }
                try
                {
                    Thread.sleep(30);
                } catch (InterruptedException e)
                {
                    // TODO Auto-generated catch block
                    e.printStackTrace();
                }

~~~~~~~~~~~

The recognise method is returning zero sized list.

Could you suggest what I have done incorrectly?

[Jonathon Hare]

I think the problem is probably your threshold being set to NaN. If you don't want to use the threshold, just use:

KNNAnnotator<DetectedFace, String, LocalLBPHistogram> knn = 
    KNNAnnotator.create(extractor, comparator, K);

[Q3Varnam]

With reference to my earlier query, I managed to get it working with the LBPH Constructor as follows

         final FaceRecognitionEngine<DetectedFace, String> engine = new LBPHistogramBaseOptions() {
                int K = 1;
                FloatFVComparison comparison = FloatFVComparison.EUCLIDEAN;
                float threshold = Float.NaN;

                @Override
                public FaceRecognitionEngine<DetectedFace, String> createRecognitionEngine()
                {
                    if (threshold == Float.NaN)
                        threshold = comparison.isDistance() ? Float.MAX_VALUE
                                : -Float.MAX_VALUE;
                    final LocalLBPHistogram.Extractor<DetectedFace> extractor = new LocalLBPHistogram.Extractor<DetectedFace>(
                            new IdentityAligner<DetectedFace>(), blocksX, blocksY, samples,
                            radius);
                    final FacialFeatureComparator<LocalLBPHistogram> comparator = new FaceFVComparator<LocalLBPHistogram, FloatFV>(
                            comparison);
                    final KNNAnnotator<DetectedFace, String, LocalLBPHistogram> knn = KNNAnnotator
                            .create(extractor, comparator, K, threshold);
                    final AnnotatorFaceRecogniser<DetectedFace, String> recogniser = AnnotatorFaceRecogniser
                            .create(knn);
                    return FaceRecognitionEngine.create(
                            fd, recogniser);
                }
            }.createRecognitionEngine();

This is resolved