Solving the over segmentation problem in applications of Watershed Transform

Abstract

Background: The Watershed Transform consists of an image partitioning into its constitutive regions. This transform is easily adapted to be used in different types of images and it allows distinguishing complex objects. However, the implementation of the Watershed Transform for very complex images actually produces over-segmentation. In this paper we propose two algorithms to solve this over-segmentation problem. Methods: We define internal markers, by algorithms based on clustering and fuzzy logic in order to join the oversegmented regions with statistical features. To define the algorithm parameters and evaluate their performance, errors against images segmented manually were measured and ROC curves were determined. Results: The results show that the proposed methods self-adapt to the different image objects characteristics. An improvement of the accuracy is obtained. Conclusions: This analysis will contribute in images segmentation where complexity of the objects is high.