基于蜂群优化或分解的二维Arimoto灰度熵阈值分割  

Two-dimensional Arimoto gray entropy image thresholding based on bee colony optimization or decomposition

作　　者：吴一全[1,2,3,4] 殷骏[1] 朱丽[1] 袁永明 

机构地区：[1]南京航空航天大学电子信息工程学院,江苏南京210016 [2]农业部淡水渔业和种质资源利用重点实验室中国水产科学研究院淡水渔业研究中心,江苏无锡214081 [3]农业部渔业装备与工程技术重点实验室,上海200092 [4]江苏省制浆造纸科学与技术重点实验室,江苏南京210037

出　　处：《浙江大学学报（工学版）》2015年第9期1625-1633,共9页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(60872065);农业部淡水渔业与种质资源利用重点实验室开放基金资助项目(KF201313);农业部渔业装备与工程技术重点实验室开放基金资助项目(2013001);江苏省制浆造纸科学与技术重点实验室开放基金资助项目(201313);农业部东海海水健康养殖重点实验室基金资助项目(2013ESHML06);江苏高校优势学科建设工程资助项目(2012);2013年研究生学位论文创新与创优基金资助项目(DZS201203)

摘　　要：现有的Arimoto熵阈值法未考虑图像目标和背景的类内灰度均匀性,为此提出基于蜂群优化和基于分解的二维Arimoto灰度熵阈值分割方法.定义Arimoto灰度熵,导出二维Arimoto灰度熵阈值法,分别利用基于蜂群优化和基于分解的方法求解最佳阈值.基于蜂群优化方法给出中间变量的快速递推公式,利用改进的人工蜂群(MABC)优化算法搜索最佳阈值,减少迭代时适应度函数中的冗余运算.基于分解方法将求解二维Arimoto灰度熵阈值法的运算转化到2个一维空间,进一步降低计算复杂度.实验结果表明:与近年来提出的3种同类方法相比,所提出方法的分割性能更优,分割后图像中目标完整、边缘纹理清晰,具有良好的抗噪性.同时,所提出的方法运行速度快,有望满足实际系统对分割的实时处理要求.The existing thresholding methods based on Arimoto entropy do not consider the uniformity of gray scale within object cluster and background cluster.A 2D Arimoto gray entropy thresholding method based on bee colony optimization or decomposition was proposed.Arimoto gray entropy was defined and a2 D Arimoto gray entropy thresholding method was derived.The method based on bee colony optimization and another method based on decomposition were adopted to find the optimal thresholds.Fast recursive formulae for the intermediate variables were given using the method based on bee colony optimization.A modified artificial bee colony（MABC）optimization algorithm was adopted to find the optimal threshold of the 2DArimoto gray entropy method.The redundant computations of fitness function in an iterative procedure could be avoided.Using the method based on decomposition,the computations of 2D Arimoto grayentropy thresholding method were converted into two one-dimensional spaces.The computational complexity was further reduced.The experimental results show that,compared with three similar methods proposed recently,the proposed methods have superior image segmentation performance and a better antinoise performance.In the segmented images,objects are completely kept,and the edges and textures are clear.Moreover,the proposed methods have high running speed and can meet the real-time processing requirement of segmentation in the actual system.

关 键 词：图像处理 阈值分割 二维Arimoto灰度熵 改进人工蜂群优化算法 分解 快速递推算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]