Image segmentation using iterative watersheding plus ridge detection

This paper presents a novel segmentation algorithm for metallographic images, especially those objects without regular boundaries and homogeneous intensities. In metallographic quantification, the complex microstructures make conventional approaches hard to achieve a satisfactory partition. We formulate the segmentation procedure as a new framework of iterative watershed region growing constrained by the ridge information. The seeds are selected by an effective double-threshold approach, and the ridges are superimposed as the highest waterlines in the watershed transform. To tackle the over-segmentation problem, the blobs are merged iteratively with the utilization of Bayes classification rule. Experimental results show that the algorithm is effective in performing segmentation without too much parameter tuning.