基于区域的活动轮廓图像分割模型的变步长优化算法

A variable step-size optimization algorithm based on active contour model

基于偏微分方程图像分割的活动轮廓模型,基本思想是将图像分割归结为最小化一个封闭曲线的能量泛函,图像分割问题实质上是一个无约束最优化问题.传统最小化算法的数值实现过程中采用固定时间步长的方法,时间步长选取较大,迭代过程容易出现震荡现象影响分割结果,而时间步长选取较小,又会减慢收敛速度.利用Wolfe-Powell线搜索方法,提出了一种变时间步长的优化算法,在迭代过程中根据搜索方向自动调整时间步长大小,有效克服了固定时间步长出现的震荡现象和收敛速度慢的问题.

The basic idea of PDE- based image segmentation based on the active contour model is to minimize the energy functional on a closed curve. Thus,the image segmentation problem is in essence an unconstrained optimization problem. The traditional minimization uses,a fix time step which may produce some problems. The large time step may lead to the oscillation of energy functional,while the small time step will slow down the convergence speed. A variable step- size method is proposed by using the Wolfe- Powell line search method. Our method will adjust time step at each iteration according to the search direction which effectively overcome the oscillation of energy functional and the defect on convergence speed.