基于自适应形状先验的快速图像分割算法

Fast image segmentation algorithm based on adaptive shape prior

针对传统Grab Cut算法在GMM迭代参数估计阶段时间复杂度较高,当图像中含有噪声或遮挡物时容易发生分割错误的问题,提出一种结合多阶抽样GMM与自适应形状先验的图像分割算法.该算法首先根据采样数定理对像素点进行均匀多阶抽样,依据样本点估计GMM参数;然后加入形状先验项约束图像分割过程,同时对形状先验约束比例采用自适应方法进行控制,获得最终分割结果.针对形状仿射变换,运用SURF与RANSAC进行处理,使本文算法更加灵活.实验表明,本文算法分割结果更加准确,效率更高.

Image segmentation method based on GrabCut has a high time complexities in the stage of estima- ting the GMM iteratively and it is prone to produce segmentation errors when the image include noise or shelter. To improve these problems,an algorithm combining GMM with muti-sampling and adaptive shape priors is pro- posed in this paper.First,the image pixels are muti-sampled based on the sampling theorem and the GMM param- eters are estimated with samples.Then the shape priors are applied to constrain the process of image segmentation and the constraint is controlled adaptively.Finally the segmentation results are obtained.This paper handles the af- fine transformation of shape by using the method of SURF and RANSAC, in order to make this algorithm flexibil- ity.The experiments show that segmentation accuracy and efficiency are improved in the algorithm.