Facial feature detection with OpenCV with eyes and mouth corners points

Question

I'm working on a face feature detection project and I do detect the eyes, nose and mouth using OpenCv withHaarcascade xml files. But, I want to have the eyes and mouth corners points and the nose center. The goal is using it to predict emotions. I found this link that shows how it works, and I need to get to this result using JAVA. Could any one help me?

Thanks in advance.

http://cmp.felk.cvut.cz/~uricamic/flandmark/

in this part we receve the face image and we drawRect on the face:

 public void drawFaces(BufferedImage image) {
    final List<PotentialFace> faces = FacialRecognition.run(image, db);
    if (faces.isEmpty()) {
      return;
    }
    Graphics2D g2 = image.createGraphics();
    g2.setStroke(new BasicStroke(2));
    currentFaces.clear();
    for (PotentialFace face : faces) {
      final Rectangle r = face.box;
      final Color c1, c2;
      final String msg;
      if (face.name == null) {
        c1 = c2 = new Color(scale(r.x, getWidth(), 255d), scale(r.y, getHeight(), 255d), 0).brighter();
        msg = "Click to tag";
      } else {
        c1 = new Color(face.name.hashCode()).brighter();
        c2 = new Color((int) (c1.getRGB() - 10*face.confidence));
        msg = String.format("%s: %f", face.name, face.confidence);
      }
      g2.setColor(c1);
      g2.drawRect(r.x, r.y, r.width, r.height);
      g2.setColor(c2);
      g2.drawString(msg, r.x + 5, r.y - 5);
      currentFaces.add(r);
    }

Answer 1

since flandmark is the C++ library which does exactly what you want (finds corner points for eyes, mouth and center/tip of nose), I think you should just look for the mechanism of how to run this library from within JAVA. flandmark itself is OpenCV independent, only examples included in this library are using it (for face detection and displaying the results).

I found some nice-looking tutorials on how to use C++ libraries from JAVA:

1. http://thebreakfastpost.com/2012/01/23/wrapping-ac-library-with-jni-part-1/