Watershed transformation: reducing the over-segmentation problem by applying noise resucer and region merger

Image segmentation is used to identify homogeneous regions in an image, it has been a subject of research for the last three decades. It is usually the first, and most difficult task for any image understanding system. Image segmentation is usually associated with pattern recognition problems and is considered the first phase of such a process. Consequently, the success of the pattern recognition process is dependent on the quality of this initial stage. Here we examine image segmentation. We describe the watershed transformation algorithm and our variation of it. We provide results for our implementation and compare then to previously published results from traditional implementations of the watershed transformation. Finally, we believe that these results substantiate the case that our modifications to the watershed provide much improved segmentation results