No-reference Stereoscopic Image Quality Assessment Using Natural Scene Statistics

We present two contributions in this work: (i) a bivariate generalized Gaussian distribution (BGGD) model for the joint distribution of luminance and disparity subband coefficients of natural stereoscopic scenes and (ii) a no-reference (NR) stereo image quality assessment algorithm based on the BGGD model. We first empirically show that a BGGD accurately models the joint distribution of luminance and disparity subband coefficients. We then show that the model parameters form good discriminatory features for NR quality assessment. Additionally, we rely on the previously established result that luminance and disparity subband coefficients of natural stereo scenes are correlated, and show that correlation also forms a good feature for NR quality assessment. These features are computed for both the left and right luminance-disparity pairs in the stereo image and consolidated into one feature vector per stereo pair. This feature set and the stereo pair's difference mean opinion score (DMOS) (labels) are used for supervised learning with a support vector machine (SVM). Support vector regression is used to estimate the perceptual quality of a test stereo image pair. The performance of the algorithm is evaluated over popular databases and shown to be competitive with the state-of-the-art no reference quality assessment algorithms. Further, the strength of the proposed algorithm is demonstrated by its consistently good performance over both symmetric and asymmetric distortion types. Our algorithm is called Stereo Quality Evaluator (StereoQUE). (C) 2016 Elsevier B.V. All rights reserved.