摘要
梯度向量流(GVF)有效解决了主动轮廓(snakes)模型初始化和凹陷区域收敛的问题,但由于其各向同性的扩散特性,使得对弱边缘和角点的捕获能力不足。因此,致力于寻求一种GVF各向异性扩散机制。通过将拉普拉斯算子进行正交分解,分析了GVF模型的法向和切向扩散作用,发现(类似)角点处的GVF场存在明显的曲率收缩和切向退化,进一步揭示了角点和弱边缘丢失的原因。在此基础上,通过对法向GVF(NGVF)模型引入新的边缘保真项和有偏的权重系数,提出一种新的外力模型。最后,通过实验对该方法的分割准确性和计算效率进行了比较分析。实验结果表明,该方法在保持一定计算优势同时,能准确地捕获弱边缘和角点。
Problems with initialization and concavity convergence for active contours, or snakes are effectively solved by Gradient Vector Flow （GVF）. However, weak edges and corners are difficult to be caught by GVF because of its isotropic diffusion characteristic. Consequently, pursuing an anisotropic diffusion GVF is essential. By orthogonally decomposing the Laplace operator, the normal and tangent diffusion of GVF was analyzed and the curvature shortening and tangent degeneration of GVF field in the corner of edge were pointed out, which further illustrated the reason of corner and weak edge losing for image segmentation. On this basis, a novel external force model was proposed by introducing a new edge preserving term and biased weight parameter into the model of GVF in the Normal direction （ NGVF）. Finally, experiments were carried out to compare and analyze the accuracy of segmentation and computing efficiency. Experimental results show that the method can accurately catch the weak edges and corners and be effectively realized.
出处
《计算机应用》
CSCD
北大核心
2016年第A02期160-164,共5页
journal of Computer Applications
关键词
梯度向量流
曲率收缩
切向退化
角点和弱边缘保持
法向梯度向量流模型
SNAKES模型
Gradient Vector Flow （GVF）
curvature shortening
tangent degeneration
corner and weak edge preserving
Normal Gradient Vector Flow （NGVF） model
snakes model
