基于概率矩阵分解的多失真图像质量评估算法

MULTI-DISTORTION IMAGE QUALITY ASSESSMENT BASED ON PROBABILISTIC MATRIX FACTORIZATION

图像质量评价(Image Quality Assessment,IQA)是视觉感知模型研究的重要分支,使用IQA算法自动化地评估图像质量有广阔的应用前景。基于概率矩阵分解(Probability Matrix Factorization)提出一种多失真图像质量评价算法,主要贡献有:提出一种图像质量评价的新思路,即利用PMF方法从失真图像估计参考图像,把失真图像和估计参考图像之间的信息损失作为图像质量的度量;构建新颖的特征向量描述这种信息损失;使用支持向量回归(Support Vector Regression,SVR)完成图像质量模型的训练。提出的算法在多个公开的图像质量评价数据库上超过了经典方法,实验结果证明该方法与人的主观质量评价具有更好的一致性。

Image quality assessment is an important branch of visual perception modeling research. Using IQA algorithm to automatically evaluate the image quality has broad application prospects. In this paper,we proposed a multiply-distorted IQA method based on probability matrix factorization. The main contributions are as follows: firstly,we came up with a new idea for IQA,which used the PMF method to estimate the reference image from the distorted image,and the information loss between the distorted image and the estimated reference version was considered as a quality measure. Secondly,we constructed a novel feature vector to describe this information loss. Finally,the IQA model was trained with SVR. The proposed method is superior to the classical IQAs in several public image quality assessment databases,which demonstrates our method has the better performance in terms of the coherence with human subjective rating.