基于结构特征的图像质量评价模型  

An image quality assessment model based on structure feature

作　　者：梁敏瑜[1] 孙权森[1] 

机构地区：[1]南京理工大学计算机科学与技术学院,江苏南京210094

出　　处：《山东大学学报（工学版）》2012年第3期52-56,共5页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(60773172);教育部博士学科点基金资助项目(200802880017)

摘　　要：由于图像的结构特征与人类视觉系统的特性密切相关,为了提高评价结果与主观视觉的一致性,提出了一种基于结构特征的图像质量综合评价模型。该模型首先提取了与图像结构密切相关的特征,然后模拟人眼对图像区域敏感程度不同,对不同的结构成分赋予不同的权值,最后结合线性回归的思想提出了一种综合评价方法。分析了不同失真图像的实验结果,表明该评价方法的正确性,同时与峰值信噪比(peak signal to noise ratio,PSNR)和均方根误差(root mean squared error,RMSE)方法进行了比较,表明本模型更准确和符合主观视觉效果。As image structure feature is closely related to the human visual system,a new method based on structure feature was proposed to improve the result consistency with a subjective visual effect.First,image structure features were extracted though experience.And then various structural components were calculated with different weights,according to different sensitivities of eyes to image regions.Finally,a new model was proposed with linear regression.Experimental results on different kinds of images showed that the new method had high efficiency and accuracy compared to the PSNR and RMSE,and had a better consistency with a subjective visual effect.

关 键 词：图像质量评价 线性回归 人类视觉系统 结构特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]