FelzenszwalbHuttenlocherSegmenter returns 1 connected component

[Anonymous]

I am just starting out with OpenIMAJ (1.3.5) and am trying out the tutorial problem 3.1.2. Exercise 2: A real segmentation algorithm which suggests using the FelzenszwalbHuttenlocherSegmenter. (And this is one of the segmenters that will probably prove relevant to the real problems I am attempting to solve).

I first tried this with a few of my own images, both as color and grayscale, and in each case the segmenter reported 1 connected component.

I then tested against the sample images and parameters on Felzenszwalb's home page and these also returned only 1 connected component.

Please provide some guidance on the proper use of this segmenter if I am doing something incorrectly.

Thank you,
David

Code below:

 public static void main( String[] args ) throws IOException {

        URL sampleImageUrl1 = new URL("http://cs.brown.edu/~pff/segment/beach.gif");        
        URL sampleImageUrl2 = new URL("http://cs.brown.edu/~pff/segment/grain.gif");

        processDefaultFelzenszwalbImage(sampleImageUrl1,  0.5f, 500f, 50);
        processDefaultFelzenszwalbImage(sampleImageUrl2,  0.5f, 1000f, 100);

    }


    public static void processDefaultFelzenszwalbImage(URL url, float sigma, float k, int minSize) throws IOException
    {

        MBFImage input = ImageUtilities.readMBF(url);

        FelzenszwalbHuttenlocherSegmenter<MBFImage> segmenter = new FelzenszwalbHuttenlocherSegmenter<MBFImage>(sigma, k, minSize);
        List<ConnectedComponent> components = segmenter.segment(input);

        int componentsCount = components.size();
        System.out.println("Component count " + componentsCount);

    }

[Jonathon Hare]

I think the problem is with your setting of the k parameter - OpenIMAJ's internal image representation (MBFImage) uses pixel values between 0 and 1, whereas Felzenszwalb's code has images where the pixels are 0-255. If you divide your k value by 255 it should work I believe.

[Anonymous]

Changing the k value as you suggest resolves the issue. Thank you for the quick and helpful response.