From: mario.drobics@scch.at   
      
   Stefan Meyer wrote:   
   > Hi,   
   >   
   > has anybody already implemented the Fuzzy C-means   
   > Algorithm for the segmentation of natural color images?   
   >   
      
   We have done this for experimental purpose. In our experiments, however,   
   we used a self-organizing map to preprocess the data. But the approach   
   should be applicable also directly. We encoded the data as a vector of   
   (x, y, r, g, b, h, l, s). hue, lightning, and saturation are not   
   necessary but can improve subsequent interpretation. x and y are used to   
   obtain clusters that are connected within the image (or at least   
   nearby). An important question when using this additional information,   
   however, is whether to use weights or not.   
      
   Finding a good initialization for c-means is always bit tricky. There   
   are various approaches to deal with that (using external knowledge,   
   crisp clustering to find the cluster centers first, k-fold crisp   
   clustering to obtain a set of potential cluster centers which are   
   clustered afterwards, recently Alexiev and Georgieva proposed an   
   approach using hough transformations).   
      
   Regards,   
   -mario-   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)