There exist two key issues in the traditional geometric active contour (GAC): only single cue based driven force and edge leakage. In this paper, we propose two new GACs, i.e., multi-Cue based GAC (MC-GAC) and local-feature based GAC (LF-GAC), to deal with these two problems respectively. In MC-GAC, a multi-cue based force is generated by introducing the customized vector flow into the model. To utilize local feature information, LF-GAC is developed, in which local edge information is considered to produce continuous edge. Further, we unify these two GACs into a novel general GAC framework. Extensive experiments showed the advantages of the proposed method
Published in:
Pattern Recognition, 2006. ICPR 2006. 18th International Conference on
(Volume:2
)
Date of Conference: 0-0 0