基于多参数和极限学习机的图像质量评价方法研究
2020-04-09孙荣荣
孙荣荣



摘 要:现有的图像质量评价算法大部分都是基于单个参数,评价不够全面。提出了基于多参数和极限学习机的图像质量评价方法,该方法融合了传统统计参数(PSNR)、基于结构相似度的参数(SSIM)和基于自然场景分析的参数(VIF和FSIM),将这些参数作为极限学习机的输入,拟合出这些特征参数和人类主观评价分值的内在关系,挖掘其内在规律。本方法和采用单独参数评价的算法进行对比分析,在TID图像质量评价库上的实验结果表明,该方法得到了比单独参数和BP方法更好的主观感知一致性。
关键词:图像质量评价;多参数;极限学习机;函数拟合
中图分类号:TP391 文献标识码:A
Research on Image Quality Assessment Method
Based on Multi-parameter and Extreme Learning Machine
SUN Rong-rong?覮
(Shanghai Institute of Measurement and testing Technology,Shanghai 201203,China)
Abstract:Present image quality assessment methods are mostly based on single parameter,but they cannot assess the image comprehensively. This paper proposes image quality assessment method based on multi-parameter and extreme learning machine. Traditional statistical parameter(PSNR),parameter based on structural similarity(SSIM) and parameter based on natural scene analysis (VIF and FSIM) are fused. These parameters are taken as the inputs to the extreme learning machine and it can excavate the relationship and inherent law between these parameters and human subjective mean opinion score. The performance of the method in the paper and method based on single parameter and BP are compared in the TID database,the experiment results demonstrate that the method in the paper achieve better subjective perceived consistency.
Key words:image quality assessment;multi-parameter;extreme learning machine;function fitting
信息化和数字化的快速发展,日常生活和工作中对数字图像处理技术的需求日趋增多,如图像传输、存储、压缩、重建等,这些技术都需要对图像的质量进行评价来优化处理算法和在同类算法中选择较优算法,所以图像质量评价越来越必要。图像质量评价有两大类方法,主观方法和客观方法,但主观方法操作过程繁琐,耗时长、费用大,且无法嵌入自动化系统中,稳定性差且很难做到实时评价,因此设计出客观的图像质量评价方法使之与人的视觉感知保持一致成为目标。基于统计的方法有均方根误差方法(MSE) 和峰值信噪比的方法(PSNR) [1],它们意义明确,直接对图像的像素值计算,但与人的主观感受有一定差距。结构相似性理论(SSIM)[2]对参考图像的亮度、对比度和结构信息进行量化,结果较好。……
