Solving the 3D watershed over-segmentation problem using the generic adjacency graph

The watershed transformation is a useful tool for the 3D segmentation. However, over segmentation have become the key problems for the conventional algorithm. This paper presents two new methods for solving these problems. The first method is to establish a generic-adjacencies graph of regions resulting from the application of watershed segmentation and to merge these regions according to a criterion of depth. The second method that works by pre-treatment uses the generic-adjacencies graph of minima to eliminate insignificant ones. In this process we have applied a hybrid criterion of depth and concavity/convexity to obtain the significant minima, these latter will subsequently pass to the watershed segmentation for a 3D object parting. The results show the effectiveness of the proposed approach. Indeed, the use of the adjacency graph allowed us to reduce processing time. Our ways permit therefore to get a fast and efficient segmentation of 3D mesh models.