Abstract: The CV model of edgeless active contours lacks a good segmentation effect for the images with blurred boundary and non-uniform background gray distribution. For this reason, the following improvements have been made to this model in this paper. Firstly, the boundary indicator function is modified and the improved boundary indicator function is integrated into the length of the CV model. Secondly, the distance regularization term of the double-well potential is introduced into the CV model to avoid re-initialization of the level set. Thus, a level set evolution equation combining gradient and region information is obtained, and the equation is solved using the finite difference method in variation method. Finally, the numerical simulation of wood bug, live knot and dead knot image is carried out. The simulation results show that the model has good segmentation effect for images with uneven background distribution.

Key words: image segmentation, boundary indication function, CV model, boundary indication function, double-well potential distance regular term

摘要： 无边缘活动轮廓CV模型对于边界模糊以及背景灰度分布不均匀的图像缺乏良好的分割效果。基于此，对该模型进行了如下改进。对边界指示函数进行修改，将改进后的边界指示函数融入CV模型的长度项中。引用双阱势的距离正则项来避免水平集重新初始化，从而得到了一个梯度与区域信息相结合的水平集演化方程，并应用变分法中的有限差分法对方程进行求解。对木材虫眼、活节和死节图像进行了数值仿真模拟，仿真结果表明该模型对背景分布不均匀的图像具有良好的分割效果。

关键词: 图像分割, 边界指示函数, CV模型, 水平集演化方程, 双阱势距离正则项