7

This is my first post here so I am sorry if my question is not clear or there is not enough information provided.

I am currently working on an Android application that could recognize faces from pictures.

My first approach was to use JavaCV and everything works fine, except the fact that face detection takes too much of time to finish!

After that, I tried to detect faces by using FaceDetector.Face. Then I used the detected faces for training my face recognizer model. There was no error found so far.

My problem is that my model could not recognize any detected face given by FaceDetector.Face. I always get -1 from the predict function. Could anybody tell what might be wrong? Thank you in advance!

This is how I crop faces after detection:

    for(int count=0;count<NUMBER_OF_FACE_DETECTED;count++)
    {
        Face face=detectedFaces[count];
        PointF midPoint=new PointF();
        face.getMidPoint(midPoint);         
        eyeDistance=face.eyesDistance();

        left = midPoint.x - (float)(1.4 * eyeDistance);
        top = midPoint.y - (float)(1.8 * eyeDistance);

        bmFace = Bitmap.createBitmap(origiImage, (int) left, (int) top, (int) (2.8 * eyeDistance), (int) (3.6 * eyeDistance));          
        bmFaces.add(bmFace);
    }

Here is a main part of training the model.

    MatVector images = new MatVector(imageFiles.length);            
    int[] labels = new int[imageFiles.length];

    IplImage img;
    IplImage grayImage;
    FaceRecognizer faceRecognizer = createLBPHFaceRecognizer(1, 8, 8, 8, binaryTreshold);
    try
    {          
        FileInputStream fstream = new FileInputStream(working_Dir.getAbsolutePath()+"/csv.txt");
        BufferedReader br = new BufferedReader(new InputStreamReader(fstream));
        String imgInfo;

        for (int i = 0; (imgInfo = br.readLine()) != null; i++)  
        {
            String info[] = imgInfo.split(";");

            String imagePath = info[0];             
            img = cvLoadImage(imagePath);
            grayImage = IplImage.create(img.width(),img.height(), IPL_DEPTH_8U, 1);
            cvCvtColor(img, grayImage, CV_BGR2GRAY);
            images.put(i, grayImage);
            labels[i] = Integer.parseInt(info[1]);;
        }
        in.close();

        //train the FaceRecognizer model         
        faceRecognizer.train(images, labels);
    }catch (Exception e)
    {
        System.err.println("Error: " + e.getMessage());
    }

Finally I recognize face with the following code:

    public static String identifyFace(IplImage grayImg)
{
    String predictedName = "";

    //identify face from the image
    int predictedLabel = faceRecognizer.predict(grayImg);

    if(predictedLabel != -1 )
    {
        predictedName = new String(idToName.get(predictedLabel));
    }
    return predictedName;
}
Peter Lawrey
  • 525,659
  • 79
  • 751
  • 1,130
user1956743
  • 81
  • 2
  • 5
  • Please don't use DataInputStream to read text http://vanillajava.blogspot.co.uk/2012/08/java-memes-which-refuse-to-die.html – Peter Lawrey Jan 30 '13 at 20:41

2 Answers2

2

This can only happen if you don't set the threshold appropriately, see the documentation:

The method to create a LBPHFaceRecognizer is:

Ptr<FaceRecognizer> createLBPHFaceRecognizer(int radius=1, int neighbors=8, int grid_x=8, int grid_y=8, double threshold=DBL_MAX)

, where:

  • threshold – The threshold applied in the prediction. If the distance to the nearest neighbor is larger than the threshold, this method returns -1.

So in the above method signature you see, that the threshold is set to DBL_MAX by default. So if you simply leave the threshold out, then it'll never yield -1. On the other hand, if you set the threshold too low, the FaceRecognizer is always going to yield -1. That said, check what you have set binaryTreshold in your code to. Finding the appropriate decision threshold for your data is a classic optimization problem, where you have to optimize for the best threshold by a given criteria (for example based on False Acceptance Rate/False Rejection Rate).

bytefish
  • 3,697
  • 1
  • 29
  • 35
  • Hi bytefish, thanks for your reply! However, my problem is face detection, not face recognition. My application takes a lot of time for detecting faces out of a picture. On the other hand, its performance on recognizing/identifying faces is quite fast. – user1956743 Jan 15 '13 at 23:12
1

I know this is really late, but try using Haar cascade classifier from JavaCV instead of facedetector from Android itself

Joshua H
  • 754
  • 8
  • 18