基于模糊聚类与Retinex理论的多相图像分割方法  

Multiphase Image Segmentation Method Based on Fuzzy Cluster and Retinex Theory

作　　者：罗群女 闵莉花 LUO Qun-nyu;MIN Li-hua(School of Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China)

机构地区：[1]南京邮电大学理学院,南京210023

出　　处：《软件导刊》2022年第11期123-129,共7页Software Guide

基　　金：国家自然科学基金项目(11671004)。

摘　　要：由于不同对象区域之间强度范围的重合,在存在灰度不均匀的情况下很难分割图像。针对这一问题,通过引入模糊隶属度函数,提出一种新的基于Retinex理论的多相图像分割模型。该方法允许每个像素点以不同隶属度同时归属于多个区域,可真实反映出图像的不确定性,并通过极小化能量泛函实现对目标物体的提取。同时在交替极小化方法的框架下,设计一种有效的算法对模型进行数值求解。实验结果表明,该模型对于灰度不均匀的真实图像和医学图像均能有效进行分割,SA指标和Dice指标的平均值分别达到0.950 6和0.914 1。相比于相关的代表性算法,Dice值提升了0.002 7~0.010 7,k值提升了0.002 9~0.011 7。Due to the overlap of intensity ranges between different object regions,it is difficult to segment the images in the presence of intensity inhomogeneity. To solve this problem,we propose a new multiphase image segmentation model based on fuzzy membership functions and Retinex theory. The method allows that each pixel can be in several regions,which reflect the uncertainty of the images. Moreover,image segmentation is achieved by minimizing the energy functional. We apply the alternating minimization method and design an effective algorithm to solve the solution of our model. Finally,the numerical experimental results are provided to verify the better performance of the proposed model for segmenting the real images and MR images with intensity inhomogeneity than other test methods,the average values of SA index and Dice index reach 0.950 6 and 0.914 1,respectively. Compared with the related representative algorithms,the Dice value is improved by 0.002 7~0.010 7,and the k value is improved by 0.002 9~0.011 7.

关 键 词：图像分割 灰度不均匀 模糊隶属度 RETINEX理论 交替极小化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]