In this paper, a new parametric active contour called self-affine snake is proposed for medical image segmentation. It integrates the wavelet transform, parametric active contour (or snake), and self-affine mapping system to keep their strengths and avoid the weak points. In more detail, it inherits wide capture range from wavelet transform and topological consistency from snake. Furthermore, it takes advantage of self-affine mapping system in several aspects including i) convergence to weak boundaries, especially, next to strong edges, ii) reconstruction of boundary openings, and iii) progress into boundary concavities. The experimental results were performed using a number of synthetic and medical images given in four sets of experiments. Self-affine snake provided comparable/superior results in terms of both solution quality and CPU time compared to a number of frequently-used active contours such as the traditional Gaussian snake, balloon, gradient vector flow (GVF), generalized GVF, active contour without edges, and region-aided geometric snake. However, its most important properties were the significant robustness against noise and reconstruction of boundary openings. Because of the valuable advantages, the proposed algorithm is an appropriate approach, particularly, for segmentation of medical images which usually suffers from noise corruption and edge uncertainty.

Keywords