基于区域划分的曲线演化多目标分割

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．

Multi-Object Segmentation Based on Curve Evolving and Region Division

This paper presentes 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, in this paper, a kind of region division procedure is introduced. In the procedure, 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.