基于Retinex理论与局部灰度信息的图像分割方法  

Image segmentation method based on Retinex theory and local intensity information

作　　者：崔强 闵莉花 石又新 CUI Qiang;MIN Lihua;SHI Youxin(School of Science,Nanjing University of Posts snd Telecommunications,Nanjing 210023,China;Northern Information Control Research Institute Group Co.Ltd.,Nanjing 211153,China)

机构地区：[1]南京邮电大学理学院,江苏南京210023 [2]北方信息控制研究院集团有限公司,江苏南京211153

出　　处：《南京邮电大学学报（自然科学版）》2021年第3期62-71,共10页Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition

基　　金：江苏省研究生科研创新计划(KYCX19_0986);国家自然科学基金(11671004)资助项目。

摘　　要：针对灰度不均匀图像分割问题,文中提出一个新的两阶段图像分割方法。第一阶段,利用结构部分和偏移场的光滑性,提出一个基于Retinex理论的变分模型,提取图像的结构部分。第二阶段,利用偏移场的局部常数特性,提出全变差的图像分割模型,对提取出的结构部分进行分割。文中提出的方法充分利用了图像的各种先验信息,从而获得了比较精确的分割结果。此外,文中给出了第一阶段模型的理论结果,并结合交替极小化方法,设计了一种有效的数值求解算法。最后的数值实验结果验证了该算法的有效性与可行性。A novel two⁃stage image segmentation method for segmenting images with intensity inhomogeneity is proposed.In the first stage,according to the smoothness of the structural part and the bias field,a Retinex⁃based variational model is proposed to extract the image structure.In the second stage,using the local constant to describe the bias part,a total variation model is proposed and the extracted structure is further segmented.The method fully considers various prior information of the image,thus obtaining a more accurate segmentation result.Furthermore,the existence of the minimizers to the variational model at the first stage is proved.Moreover,by using the alternating minimization method,an efficient algorithm is designed to solve numerically the solution of the model.Experimental results validate the effectiveness and the feasibility of the method.

关 键 词：图像分割 灰度不均匀 RETINEX理论 局部常数 交替极小化 

分 类 号：TN911.73[电子电信—通信与信息系统]