基于区域划分的曲线演化多目标分割  

Multi-Object Segmentation Based on Curve Evolving and Region Division

作　　者：杨莉[1] 杨新[1] 

机构地区：[1]上海交通大学图像处理与模式识别研究所,上海200030

出　　处：《计算机学报》2004年第3期420-425,共6页Chinese Journal of Computers

基　　金：国家自然科学基金 ( 6 0 0 72 0 2 6 )资助

摘　　要：该文在小波变换的多分辨率框架下建立了一种基于曲线演化的多目标分割算法 ,并且目标分割由两步实现 :(1 )区域划分 ,将图像域分为多个子区域 ;(2 )在各子区域中 ,采用基于简化的Mumford Shah模型的曲线演化方法进行分割 ,从而实现了多个 (不局限于一个 )不同平均灰度目标的分割 .由于算法建立在区域划分和CV方法的基础上 ,因而对受噪声影响大、边缘模糊的多个不同质区域仍能得到正确的分割 .同时算法从多方面提高了曲线演化速度 .This paper presents a multi-object segmentation method based on CV method, a level set method based on Mumford-Shah model, which is approved powerfully for segmentation on high-noise and smooth-edge images. In order to extend the CV segmentation method to the multiple objects images with more than two different mean intensities, a kind of region division procedure is introduced. Image domain is firstly divided into multiple sub-regions, and then curve evolution is performed in each sub-region. In addition, necessary definitions and theorems are proposed and proved mathematically. Theoretical analysis and computer simulation data illustrate that the procedure introduced in this paper can effectively and robustly deal with images with multiple non-overlapping objects with different means, that is, no longer restricted just to images with only two distinct mean intensities. Moreover, the segmenting process is speeded up.

关 键 词：图像分割 计算机视觉 图像处理 区域划分 曲线演化 多目标分割算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]