基于支持向量回归的无参考模糊和噪声图像质量评价方法

基于支持向量回归(SVR)和图像奇异值分解,提出了一种新的无参考(NR,no-reference)模糊和噪声图像质量评价(IQA)方法。首先通过对待评价图像进行高斯低通滤波生成再模糊图像,然后分别对它们进行奇异值分解并计算奇异值的改变量,最后使用奇异值的改变量作为SVR的输入,训练预并测得到图像的质量评分。在3个公开的模糊和噪声数据库上的实验结果表明,新方法预测得分与主观得分有较好的一致性,获得了较好的评价指标;对于模糊失真类型和噪声失真类型,在LIVE2数据库上的性能评价指标斯皮尔曼等级相关系数(SROCC)分别达到0.961 3和0.965 9。

No-reference quality assessment algorithm for blur and noise images using support vector regression

In this paper,we propose a new no-ref erence image quality assessment (IQA) algorithm for blur and noise images using support vector regression (SVR) and singular value decomposition. The algorithm is composed of three steps.First,a re-blurred reference image i s produced by using Gaussian low-pass filter for a test image.Then we do singular value decomposi tion to them and calculate the change of their singular values.Thirdly,we train the support vector regres sion by using change of singular values and predict image quality score.Experimental results on three o pen blur and noise databases show that the proposed algorithm is more reasonable and stable than ot her methods.It has high correlation with human judgments and obtains a better evaluation index.So the proposed method is appropriate for no-reference blurred and noise image quality assessment.For the blur and noise distortion types,the performance indices of Spearman rank correlation coef ficient (SROCC) on the LIVE2database can reach 0.9613and 0.9659,re spectively.