圆形直方图线性化的高精度高适应性多阈值分割方法  

Multi-threshold Segmentation Method with High Accuracy and Adaptability Using Circular Histogram Linearization

作　　者：黄聪[1,2] 邹耀斌 孙水发[1,2] HUANG Cong;ZOU Yaobin;SUN Shuifa(Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering,China Three Gorges University,Yichang 443002,Hubei,China;College of Computer and Information Technology,China Three Gorges University,Yichang 443002,Hubei,China)

机构地区：[1]三峡大学湖北省水电工程智能视觉监测重点实验室,湖北宜昌443002 [2]三峡大学计算机与信息学院,湖北宜昌443002

出　　处：《计算机工程》2024年第1期259-270,共12页Computer Engineering

基　　金：国家自然科学基金(61871258)。

摘　　要：针对已有彩色图像多阈值分割方法存在的分割精度不高、分割适应性较差等问题,提出一种圆形直方图线性化的高精度高适应性多阈值分割方法。在对输入彩色图像进行超像素预处理后,构建累积分布方差最大化准则,将圆形直方图截断后延展为线性直方图。在线性直方图上,结合Tsallis熵和类间方差构建一个新的多阈值分割目标函数。引入麻雀搜索算法,快速求解多阈值分割目标函数得到最优阈值进行阈值分割。在8幅合成图像和500幅真实世界图像上将提出方法和9种不同的彩色图像分割方法进行全面比较,在峰值信噪比(PSNR)、结构相似性(SSIM)、特征相似度(FSIM)、概率兰德指数、全局一致性误差(GCE)、信息差异6个量化评价指标上的综合实验结果表明,提出方法在计算效率方面与比较方法大致持平,但在分割精度和分割适应性方面明显优于比较方法,在PSNR、SSIM、FSIM和GCE等评价指标上分别以19.95 dB、0.80、0.94和0.16取得最优结果。To address the problems of low segmentation accuracy and poor adaptability of existing multi-threshold segmentation methods for color images,a multi-threshold segmentation method with high accuracy and adaptability based on circular histogram linearization is proposed.After superpixel preprocessing of the input color image,the method first constructs the cumulative distribution variance maximization criterion,based on which the circular histogram is truncated and extended into a linear histogram.Thereafter,a new multi-threshold segmentation objective function is constructed by combining the between-class variance and Tsallis entropy on the linear histogram.Finally,the Sparrow Search Algorithm(SSA)is introduced to quickly solve the multi-threshold segmentation objective function to obtain the optimal threshold.On eight synthetic images and 500 real world images,the proposed method is comprehensively compared with nine different color image segmentation methods.The comprehensive experimental results on six quantitative evaluation indicators,such as Peak Signal-to-Noise Ratio(PSNR),Structural Similarity Index Measure(SSIM),Feature Similarity Index Measure(FSIM),Probabilistic Rand Index(PRI),Global Consistency Error(GSE),and Variation of Information(VI),show that,the proposed method is approximately equal to the compared method in computational efficiency,but it is significantly better than the compared nine methods in segmentation accuracy and adaptability.The proposed method is ranked first in terms of PSNR(19.95 dB),SSIM(0.80),FSIM(0.94),and GSE(0.16).

关 键 词：多阈值分割 圆形直方图 累积分布方差 自适应Tsallis熵 麻雀搜索算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]