基于混合高斯模型和相似度的阈值分割

摘 要:阈值分割法在图像分割中是一种最为简单和有效的方法，然而如何选取合适的阈值实现有效的分割，目前还没有统一的方法。本文结合信息论与图像的空间信息，提出一种新的阈值优化算法。首先建立图像的混合高斯分布（GMM），然后利用强度的类不确定性和相似度函数特性定制出目标函数，优化出局部阈值，从而获得高效的分割效果。实验结果表明，与大津法（OSTU）相比，本文提出的算法能够成功分割出模糊的边界,并且能够将图像中的各个组织有效的分割出来。

Based on GMM model and similarity of the threshold segmentation

Abstract: Threshold segmentation method in image segmentation is one of the most simple and effective method, but how to select appropriate threshold to achieve effective segmentation is still not unified method at present. Based on information theory and image space information, this paper puts forward a new threshold optimization algorithm. First establish image mixed Gaussian distribution (GMM), and by using the intensity of the class uncertainty and similarity function characteristic custom out the target function, optimize the local threshold value, so as to achieve efficient segmentation effect. The experimental results show, compared with the OSTU, the proposed algorithm can successfully segment fuzzy boundary and image of each effective organization segmentation out.