基于张量域感知特征的无参考高动态范围图像质量评价

No-reference Image Quality Assessment of High Dynamic Range Image Based on Tensor Domain Perceptual Features

针对传统的无参考图像质量评价方法不能直接用于高动态范围(HDR)图像的质量评价问题,提出一种基于张量域感知特征的无参考HDR图像质量评价方法.首先通过张量分解得到同时含有亮度失真和色度失真的张量子带图;然后采用自回归(AR)模型来模拟人脑对张量子带图的感知预测,得到张量子带图的感知预测图像;最后采用AR系数表征HDR图像的张量域感知预测特征,结合张量子带图及其感知预测图的动态范围、亮区域占比的特征,使用支持向量回归模型进行回归映射得到图像的客观质量评价分数.在Nantes和EFPL这2个公开的HDR图像库上的实验结果表明,该方法与主观感知具有很好的一致性,评价指标SROCC和PLCC的值均超过0.93, RMSE分别为0.3047和0.3771.

For the traditional no-reference image quality assessment method cannot be directly applied to the quality evaluation of high dynamic range(HDR)images,a no-reference HDR image quality assessment method based on tensor domain perceptual features is proposed.Firstly,tensor decomposition is used to obtain the tensor sub-bands with luminance distortion and chrominance distortion.Then,the Auto-Regressive(AR)model is adopted to simulate the process that human brain adopts to predict the tensor sub-bands and obtain the perceptual prediction image of the tensor sub-bands.Finally,the AR coefficients are employed to represent perceptual predictive characteristics of HDR image in tensor domain,and the quality score of the HDR image is obtained via support vector regression model,combining with the dynamic range and proportion of brighter areas of the tensor sub-bands and the perceptual prediction image.Experimental results tested on two public HDR image databases of Nantes and EFPL show that the proposed method can achieve high consistent alignment with subjective assessment.The performance indices of SROCC and PLCC are all above 0.93,RMSE are 0.340 7 and 0.377 1,respectively.