Watershed-based region merging using conflicting regions

In this paper, we present a method of watershed-based region merging using 'conflicting regions' for segmentation of gray level images. It is obvious that both regions and edges in an image give important clues to segmentation in our visual system. So our method uses information from both regions and edges properly. We first obtain initial segments by applying watershed transformation to the image gradient magnitude and then find 'conflicting regions' by using the edges. After that, 'conflicting regions' and region homogeneity guide the iterative merging process. 'Conflicting regions' give a good starting point of merging and make it unnecessary to define termination criterion in advance. Since they serve as seeds, processing time is also inexpensive. The experimental results show that segmentation is visually reasonable.