基于贝叶斯学习的数字化图像多阈值分割方法  

Digital Image Multi-threshold Segmentation Method Based on Bayesian Learning

作　　者：何宇新 陈勇 王友元 HE Yu-xin;CHEN Yong;WANG You-yuan(China Southern Power Grid Digital Enterprise Technology(Guangdong)Co.LTD,Guangzhou Guangdong 510663,China;Central South University School of Mathematics And Statistics,Changsha Hunan 410083,China)

机构地区：[1]南方电网数字企业科技(广东)有限公司,广东广州510663 [2]中南大学数学与统计学院,湖南长沙410083

出　　处：《计算机仿真》2024年第5期226-230,269,共6页Computer Simulation

基　　金：南网数研院人力资源管理人才数字化建设项目(0002200000085847)。

摘　　要：经典Otsu分割算法精度高,适应度强,但由于其分割稳定度较差,计算效率较低,在实际使用时具有一定的应用局限。为提升Otsu图像分割的稳定度与计算效率,提出一种基于贝叶斯网络结构学习算法对搜索节点进行优化,提高整体计算效率。首先通过在BN层通过独立性测试,构架初始种群BN_0,并利用GR算法不断修正进化方向;然后构建MSWT树网解决种群扩张时多父节点导致算法收敛速度低的问题;接着采用ACO算法规划个体转移禁忌表,通过结合个体转移概率,完成最优阈值规划;最后构建OEA-Otsu数字图像多阈值分割模型,通过解析适应函数进行并利用最优Gbest结构规划完成图像多阈值分割输出。仿真结果表明,较其它基线优化算法相比,经OEA优化后的图像多阈值分割模型,FSIM指标整体提高了5.91%,SSIM指标至少提升了2.33%,PSNR指标分析在实验数据上分别增加了12.76%、10.74%以及12.48%。即提出的OEA-Otsu模型有效的提高了图像分割的稳定度差与计算效率。The classical Otsu segmentation algorithm has high accuracy and strong fitness,but it has certain application limitations in practical use because of its poor segmentation stability and low computational efficiency.In order to improve the stability and computational efficiency of Otsu image segmentation,this paper proposes a Bayesian network structure learning algorithm to optimize the search nodes and improve the overall computational efficiency.Firstly,through the independence test in the BN layer,the initial population BN_0 was constructed,and the GR algorithm was used to continuously modify the evolutionary direction;then the MSWT tree network was constructed to solve the problem of low convergence rate caused by multiple father nodes when the population expands;then the ACO algorithm was used to plan the individual transition tabu list,and the optimal threshold planning was completed by combining the individual transition probability;finally an OEA-Otsu digital image multi-threshold segmentation model was constructed,and the image multi-threshold segmentation was carried out by analyzing the fitness function and using the optimal Gbest structure planning.The simulation results show that compared with other baseline optimization algorithms,the FSIM index of the image multi-threshold segmentation model optimized by OEA is improved by 5.91%,the SSIM index is improved by at least 2.33%,and the PSNR index analysis is increased by 12.76%,10.74%and 12.48%respectively in the experimental data.That is to say,the proposed OEA-Otsu model effectively improves the stability and computational efficiency of image segmentation.

关 键 词：贝叶斯学习 图像多阈值分割 算法 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]