一种基于变分水平集的红外图像分割算法

A Segmentation Algorithm of Infrared Image Based on Variational Formulation Level Set Model

红外图像大都存在边缘模糊或离散状边缘的特点,并且图像的先验知识较少,因此红外图像的分割是比较困难的。针对这种情况,该文提出了一种基于图像全局信息并且不需要重新初始化的变分水平集红外图像分割方法,不考虑图像边缘梯度的影响,将图像全局信息作为外部能量项,在很大程度上克服了边缘模糊时过分割的情况。同时通过引入内部变形能量约束水平集函数逼近符号距离函数,省去了重新初始化水平集函数的过程,简化了计算,减小了因重新初始化水平集函数带来的误差。将算法应用在红外图像的分割中,验证了算法的有效性。

Infrared images always have little priori-information and blurry boundaries or even with discontinuous boundaries, and therefore the segmentation of infrared image is very difficult, A variational formulation level set model based on the image global information is obtained, which can detect contours both with and without gradient based on global information and eliminate the re-initialization procedure. The stopping term does not depend on the boundary gradient of the image, which overcomes greatly the ＂over segmentation＂. This algorithm forces the level set function to be close to a signed distance function, and therefore reduces the number of iterations and the error of re-initialization. The method is used in the segmentation of an infrared image, and the experiment results reveal the effectiveness of the algorithm listed in this paper.