如何在Javacv中获得人脸相等范围

Question

I'm using Javacv 1.1. have used an image of John lennon to test a Face recognition, that contains a data base with 3 images of Cristiano Ronaldo and 3 of John Terry. Just to verify that my program not find john lennon in the data base I used a photo from his.

But when program predicts who was the guy in the image, shows me that it been Cristiano Ronaldo as the most similar guy.

I need the distance between the input image and most similar training image to check if is the enough similar to tell me who is, or if is not enough tell the guy in there is not known.

Another question: Any tip for improve a correctly match?

Answer 1

This was my solution ( tested only for Android platform), assuming you use eigenfaces:

public boolean verifyPerson (Mat matResized) {
        CvArr eigenvectors = faceRecognizer.getMat("eigenvectors");
        CvArr averageFaceRow = faceRecognizer.getMat("mean");
        IplImage img1D =MatToIplImage(matResized.reshape(1,1),-1, -1);
        // Project the input image onto the eigenspace:
        CvMat projection = subspaceProject(eigenvectors, averageFaceRow, img1D);
        // Generate the reconstructed face back from the eigenspace:
        CvMat reconstructionRow = subspaceReconstruct(eigenvectors, averageFaceRow, projection);
        CvMat reconstructionMat = new CvMat();
        cvReshape(reconstructionRow, reconstructionMat, 1, matResized.height()); 
        Mat reconstructedFace = CvMatToMat(reconstructionMat);
        // Calculate the L2 relative error between the 2 images.
        double errorL2 = Core.norm(matResized, reconstructedFace, CV_L2);
        // Scale the value since L2 is summed across all pixels.
        double uncertainty = errorL2 / (double)(matResized.rows() * matResized.cols());
        boolean verified;
        if (uncertainty < MaxPersonUncertainty) {
            verified = true; 
            }else{
            verified = false; // Unknown person.
            }
        return verified;
    }