Watershed Segmentation with Chamfer Metric Vasily Goncharenko 1 and Alexander Tuzikov 2 1 National Center of Information Resources and Technologies, National Academy of Sciences of Belarus, Akademicheskaja 25, 220072 Minsk, Belarus vasily@mpen.bas-net.by 2 United Institute of Informatics Problems, National Academy of Sciences of Belarus, Surganova 6, 220012 Minsk, Belarus tuzikov@newman.bas-net.by Abstract. Watershed transformation is introduced as a computation in image graph of a path forest with minimal modified topographic distance in (R + ) 2 . Two algorithms are presented for image segmentation that use a metric defined by a unit neighborhood as well as a chamfer (a, b)- metric. The algorithms use ordered queues to propagate over image pixels simulating the process of flooding. Presented algorithms can be applied to gray-scale images where objects have noticeable boundaries. 1 Introduction Watershed transformation of gray-scale images often results in better segmenta- tion and contour detection outcomes in comparison with other methods. Image segmentation by watershed transformation belongs to the region growing meth- ods that combine pixels according to similarity of their properties relative to the properties of their local neighbors. This method works good for images with objects characterized by brightness or color characteristics rather than texture features. Watershed transformation of gray-scale images was firstly described by S. Beucher and C. Lantu´ejoul [1] in their paper devoted to contour detection of objects on metallographic pictures. The authors adopted geographic termi- nology for describing the contour detection process, based on the disclosure of the areas with the greatest absolute gradient values. The image was presented as a topographic surface the drops of water fall on, stream down and come into local minima. Each local minimum has its own set of points of the surface named catchment basin. If a drop falls on a point belonging to a catchment basin, it will come down into a corresponding local minimum. Several catchment basins may intersect - their common points form watersheds. Formal construction of watershed points was based on the detection of points equidistant from different catchment basins lying on a given level λ. In a later work S. Beucher [2] considered two groups of watershed trans- formation algorithms. The first group contained algorithms which simulate the flooding process (or immersion into water). An algorithm based on morphological operations was taken as an example. The second group was made of procedures detecting watershed points directly. A. Kuba, L.G. Ny´ ul, and K. Pal´agyi (Eds.): DGCI 2006, LNCS 4245, pp. 518–529, 2006. c  Springer-Verlag Berlin Heidelberg 2006