二维最优进化图像分割算法  

2D optimal evolution algorithm for image segmentation

作　　者：吴伟文[1] 林正春 王知衍[1] 王成[3] 梁鹏[4] 

机构地区：[1]华南理工大学计算机科学与工程学院,广东广州510006 [2]佛山标准技术研究院研发部,广东佛山528000 [3]仲恺农业工程学院计算机科学与工程学院,广东广州510225 [4]广东技术师范学院计算机科学学院,广东广州510665

出　　处：《微型机与应用》2012年第13期38-41,45,共5页Microcomputer & Its Applications

基　　金：国家质检总局科技计划项目(2010QK094)

摘　　要：针对一维阈值分割只考虑图像的灰度级而不考虑像素的空间信息且需要目标函数的问题,根据最优进化图像阈值分割算法的基本思想,提出了一种无目标函数的二维图像阈值分割算法框架(2D-OEA)。2D-OEA将每个图像二维信息向量看作一个染色体,假设最优进化方向存在,建立进化方向更新模型;然后定义了染色体编码规则,通过简单随机采样初始化种群,再对种群进行交叉变异运算、适值计算、选择和阈值修正,得到稳定的最优二维阈值。分别从理论和实验分析了假设和模型的合理性。实验结果表明,假设和进化方向更新模型合理,2D-OEA快速、稳定且有效,分割结果优于OEA。1D image thresholding algorithm only considers the distribution of the pixel grayscales, but ignores the correlation between different gray levels, and needs an objective function. To solve these problems, a 2D optimal evolution algorithm（2D-OEA） without an objective function for image thresholding is proposed based on optimal evolution algorithm. The 2D-OEA regards every. 2D information vectors as chromosome. Assuming the optimal evolution direction exists, the updating model of evolution direction is established. Then it defines the chromosomes＇ coding rules, initializes the group by simple-random-sampling, selects chromosomes to crossover and mutate, calculate the fitness values, produces a new population by the selection mechanism and modiies the threshold to obtain a stable 2D optimal threshold. The rationalities of the assumption and the updating model have been analyzed in this paper. The experimental results show that the assumption and the updating model are proper, 2D-OEA is a fast, robust and effective algorithm, and it is better than OEA.

关 键 词：阈值分割 图像分割 遗传算法 二维直方图 灰度级 

分 类 号：TP931.4[自动化与计算机技术]