逐层迭代的多尺度水平集红外图像分割算法  

Multiscale level set segmentation algorithm for infrared image using iterative tier-by-tier

作　　者：梅雪[1,2] 林锦国[1] 夏良正[2] 

机构地区：[1]南京工业大学自动化学院,南京210009 [2]东南大学自动化学院,南京210096

出　　处：《东南大学学报（自然科学版）》2008年第A02期96-100,共5页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(60805002);江苏省“六大人才高峰”资助项目(07E024);江苏省高新技术重点实验室资助项目(BM2007201)

摘　　要：针对应用传统水平集方法进行红外图像分割易产生误分割且运算量大,提出了一种多尺度水平集分割算法,该算法利用小波变换将图像分解成多尺度子图像,在子图像上进行图像分割,采用插值法将粗尺度上曲线演化结果投影到细尺度上作为初始轮廓线,逐层分割,并根据图像噪声的大小,确定演化模型中的正则项参数,使噪声得到有效抑制,获得准确的分割结果.仿真实验表明,该方法不仅分割效果好,而且具有抗噪性强和运算速度快的特点.Infrared images always have blurry boundaries with great noise, and mis-segmentation may occur when traditional level set method is used in infrared image segmentation. An effective multiscale level set algorithm is thus proposed to solve this problem. Wavelet transform is used to get multiscale images. Then segmentation is completed firstly in finer scale sub-image, and the interpolation method is used to pass the evolving result from the coarser scale to the finer one, iterating tierby-tier. This algorithm estimates the parameter of the regularizing term of the evolving model adaptively based on the variance of the noise. The experimental results show that this method not only efficiently implements the image segmentation, but also suppresses the noise greatly and decreases the cost time.

关 键 词：水平集 红外图像分割 多尺度分析 小波变换 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]