基于混沌粒子群优化的倒数熵阈值选取方法

Thresholding based on reciprocal entropy and chaotic particle swarm optimization

基于信息熵的方法是一类重要的阈值选取方法,但现有的最大熵方法存在无定义值问题。为此,提出了基于倒数熵的阈值选取方法。首先给出了倒数熵的定义及一维阈值选取方法,导出了基于二维直方图区域直分及区域斜分的倒数熵阈值选取算法公式;然后考虑到二维倒数熵分割运算量较大,提出利用混沌小生境粒子群算法来寻找最优阈值,避免了算法早熟,提高了搜索精度和算法效率。实验结果表明:二维倒数熵阈值选取的斜分方法在抗噪性和运算时间上优于直分方法;而与基于粒子群优化的二维最大熵方法相比,本文提出的基于混沌小生境粒子群优化的二维倒数熵斜分法在运行时间上降低了约40%,分割效果更佳。

Method based on information entropy is a class of important threshold selection method,but the existing maximum entropy method has the problem of undefined value. Thus,the threshold selection method based on reciprocal entropy is proposed. Firstly,reciprocal entropy is defined and one-dimensional threshold selection method is given. The algorithm formulae for reciprocal entropy threshold selection based on two-dimensional histogram vertical segmentation and oblique segmentation is derived. In view that the computational burden of two-dimensional reciprocal entropy segmentation is large,Niche chaotic mutation particle swarm optimization（NCPSO） is adopted to find the optimal threshold. It avoids algorithm premature and improves searching accuracy and efficiency. The experimental results show that oblique segmentation method of two-dimensional reciprocal entropy has advantages over vertical segmentation method both on anti-noise and running time. Compared with two-dimensional maximum entropy method based on particle swarm optimization（PSO）,two-dimensional reciprocal entropy oblique segmentation method based on NCPSO is reduced by 40% or so in processing time,and achieves superior segmentation quality.