基于广义梯度矢量流模型图像分割算法的研究

Research on image segmentation algorithm based on generalized gradient vector flow model

梯度矢量流模型(GVF)是引导主动轮廓模型形变的有效外力,但不能分割细长凹陷区域,因此提出广义梯度矢量流(GGVF)模型。GGVF模型将GVF模型中的常数系数替换为两个变化的权重函数,虽然拥有较好的细长凹陷收敛能力,但仍然在进入复杂凹陷、保护弱边缘、抵抗噪音方面存在局限性。为了解决这些问题,在GGVF模型的基础上做了3点改进:1)在构建边缘图时引入双边滤波器用于平滑噪音;2)用散度算子替代拉普拉斯算子实现更好的凹陷收敛能力;3)添加方向约束函数用于保护弱边缘。实验结果表明,该算法具有较好的性能,且相对于传统GGVF模型,召回率提高15.6%,F1值提高8.7%。

The gradient vector flow(GVF)is an effective external force to deform the active contours.Due to the difficul⁃ty in evolving into long and thin indentations(LTIs),the generalized GVF(GGVF)has been proposed by replacing the constant coefficients in the GVF model with two varying weight functions.The GGVF model has better ability of LTIs convergence,but it still has limitations in entering complex concavities,suppressing noise and protecting weak edges.To address these problems,we made three improvements based on GGVF model:1)Incorporate a bilateral filter when con⁃structing a novel edge map to suppress noise,2)Replace the Laplacian operator with the divergence operator to achieve better performance of concavity convergence,and 3)Employ an orientation constraint function to preserve weak edges.Experiments show that this method has good performance,and has increased about 15.7%and 8.7%in recall and F1 measure compared with the traditional GGVF model.