Abstract: A new improvement based on Gradient Vector Flow(GVF) model is put forward, in order to solve the problem brought about by linear diffusion that the traditional GVF model will often blur the boundary of object map and weak boundary leaking. This improvement adopts eight directional non-linear diffusion to keep the boundary of object map, and makes use of a new coefficient of fidelity term which have a faster descent speed to strengthen Snake to segment the depression part. Theoretical analysis and experimental result show that, this new improvement has better ability to segment the depression, and has better robustness concerning weak boundary leaking.

Key words: image segmentation, Snake model, Gradient Vector Flow(GVF) model, non-linear diffusion

摘要： 针对传统梯度矢量流(GVF)模型各向同性扩散分割图像时所导致的模糊边缘以及弱边界处的泄漏问题，提出一种新的GVF模型，该模型采用8个方向各向异性扩散策略以保持目标边界，并使用具有较快下降速度的保真项系数来增强Snake进入凹陷部分的能力。理论分析和实验结果表明，新方法能较准确地分割出目标凹陷部分，对于弱边界泄漏具有更强的鲁棒性。

关键词: 图像分割, Snake模型, GVF模型, 各向异性扩散

CLC Number: 

• TP391.41