Segmentation of complex shapes by adaptive energy forces

Abstract

The classical gradient vector flow technique, to some extent has the ability to catch up dynamic topological changes, and therefore to extract complex shapes. Due to the reliance on the detected edges and the corresponding strength, the snake may be obstructed to rest on the ideal contours. To remedy these two deficiencies, a new deformable model is proposed in this paper. The idea is to improve the energy function by consistently reducing the Euclidean distance between the initial centroid and the estimated one of the snake. This is achieved by applying the mean shift for estimating the varied centroid of the snake during the iteration, which indicates the balance point of the overall forces. Experimental results show favorable performance of the proposed approach.