Active contour model driven by linear speed function for local segmentation with robust initialization and applications in MR brain images

摘要

Active contour model (ACM) has been widely used in image segmentation, but the local region based ACM suffers from the sensitivity of the curve initialization, which means that the segmentation can be influenced by the initialized contour greatly. In this paper, we propose a novel local region based ACM. Firstly, we analyze the reason for the sensitivity of the curve initialization, i.e., the blind region and the false edges. Secondly, we propose a novel local region-based linear speed function, in which, the additive factor can solve the blind region and false edge problems, and the multiplicative factor can further improve the additive factor in solving false edge problem. Thirdly, we incorporate the proposed linear speed function into the local approximated signed distanced function based local segmentation framework. In the proposed method, we only need to select one point anywhere inside the object for initialization, which is very convenient for interactive segmentation. Experiments on synthetic and Magnetic Resonance (MR) brain images demonstrate the robustness of the initialization over the ACM driven by the classic local region-based intensity energy, ACM driven by local and global intensity energy and ACM driven by contrast constrained local intensity fitting energy.