Improving the original Dual-T-Snakes model

The dual-t-snakes model plus dynamic programming (DP) techniques is an efficient methodology for boundary extraction and segmentation of 2D images. However, the original proposal of the dual-t-snakes suffers from difficulties for parameters choice and instabilities due to the normal force definition. Also, the proposed method to push a snake way from a local minimum is not efficient for noise images. In this paper we address these limitations. Firstly, we reduce the number of parameters without affecting the basic features of dual-t-snakes. Next, we propose a new normal force definition which maintains the desired features of a balloon-like one but is more stable and gives better performance. Then, we propose other methods to avoid local minima based on mesh resolution, image statistics and a new region growing technique. We demonstrate these methods for artificial and cell images. Finally, in the future works, we figure out how to apply dual-t-snakes plus Viterbi together with non-parametric multiscale methods