Weighted kernel mapping model with spring simulation based watershed transformation for level set image segmentation

Abstract

This paper proposes a novel active contour model called weighted kernel mapping (WKM) model along with an extended watershed transformation (EWT) method for the level set image segmentation, which is a hybrid model based on the global and local intensity information. The proposed EWT method simulates a general spring on a hill with a fountain process and a rainfall process, which can be considered as an image pre-processing step for improving the image intensity homogeneity and providing the weighted information to the level set function. The WKM model involves two new energy functionals which are used to segment the image in the low dimensional observation space and the higher dimensional feature space respectively. The energy functional in the low dimensional space is used to demonstrate that the proposed WKM model is right in theory. The energy functional in the higher dimensional space obtains the region parameters through the weighted kernel function by utilising mean shift technique. Since the region parameters can better represent the values of the evolving regions due to the better image homogeneity, the proposed method can more accurately segment various types of images. Meanwhile, by adding the weighted information, the level set elements can be updated faster and the image segmentation can be achieved with fewer iterations. Experimental results on synthetic, medical and natural images show that the proposed method can increase the accuracy of image segmentation and reduce the iterations of level set evolution for image segmentation.