无参考混合失真图像质量评价方法

No-Reference Quality Assessment Method for Multiply Distorted Images

目的为了解决图像在实际处理过程中产生混合失真的情况,研究无参考混合失真图像质量评价方法。方法首先利用空域-频域信息熵和奇异值建立无参考混合失真图像失真类型判别模型,然后再根据不同的混合失真类型,分别提取多维空间统计特征、奇异值改变量和空域-频域信息熵等3种不同的图像信息特征,建立无参考混合失真图像质量评价模型,并选取最佳无参考图像质量评价(NR-IQA)模型得到分数。结果该方法能100%判别混合失真类型,对于模糊噪声、模糊压缩混合失真在LIVE混合失真图像数据库(LIVEMD)上的斯皮尔曼等级相关系数(SROCC)最高分别能达到0.9874和0.9916,具有很好的主观一致性。结论实验结果表明,该无参考混合失真图像质量评价方法与人眼视觉感知具有很好的主观一致性。

The work aims to study the no-reference quality assessment method for multiply distorted images, in order to solve the problem that the images are simultaneously distorted by multiple types of distortions in practice. First of all, the discriminative model of no-reference multiply distorted images was established by spatial and spectral entropies and singular values. Then, according to the different multiple types of distortions, the no-reference quality assessment（NR-IQA） model for multiply distorted images was established by respectively extracting such three types of different image information features as multi-dimensional spatial statistical feature, singular value variation and spatial and spectral entropies. Moreover, the scores were obtained by selecting the optimal NR-IQA model. The discrimination rate of such method for the multiple distortion types could reach up to 100%. For the multiply distorted images, involving fuzzy noise and fuzzy compression, the spearman rank-order correlation coefficient（SROCC） on the LIVEMD database were 0.9874 and 0.9916, respectively, which provided good subjective consistency. The experiment results show that, such no-reference quality assessment method for multiply distorted images is in good subjective consistency with the ocular perception.