Color image segmentation using PDE-based regularization and watershed transformation

We propose an extension of watershed segmentation to color images. PDE-based regularization is considered prior to the extraction of watershed contours in order to avoid the inherent oversegmentation problem. Several PDEs are implemented and tested in terms of their efficiency as a preprocessing block of watershed segmentation. We have also implemented the fast watershed algorithm of L. Vincent and P. Soille (see IEEE Trans. on PAMI, vol.13, no.6, 1991) in concatenation with region merging to get a true segmentation of color images. The results prove to be satisfactory.