No reference stereo video quality assessment based on motion feature in tensor decomposition domain

A no reference stereo video quality assessment method based on motion features extracted in tensor decomposition domain is proposed. Tensor decomposition is used to reduce dimension of color, view and time of stereo video, and motion information maps containing time-varying information of inter-views and intra-views are obtained. Statistical features such as generalized Gaussian distribution (GGD), asymmetric GGD, spatial entropy, spectral entropy associated with two views, and spectral entropy related to depth perception of stereo video, are extracted. Random forest is utilized to establish relationship between stereo video quality and the extracted features. Experimental results on NAMA3DS1-COSPAD1 database demonstrate that the proposed method achieves good performance on JP2K, resolution reduction, sharpening and their combination distortions, Pearson linear correlation coefficient (PLCC) values of these types of distortions are higher than 0.97, while for H.264 distortion the PLCC value is 0.8850, which means that the proposed metric is consistent with human visual perception.