Using 3D Structural Tensors in Quality Evaluation of Stereoscopic Video

Abstract

Recent advancements in 3D imaging, coding, compression, storage, transmission, and error concealment techniques enable wide usage of 3D video/image applications. Video quality assessment plays a major role in improving the perceived quality at the receiver side, since such information can be used at the transmitter or in the different network nodes for system optimization `on-the-fly'. Most of the objective quality metrics in use are Full-Reference (FR), requiring the original video sequance for comparison. In the case of quality assessment for stereoscopic video, both left and right views need to be considered. In this paper, we introduce a novel Reduced-Reference (RR) quality metric for stereoscopic video using 3D structural tensors, based on the fact that the Human Visual System (HVS) is more sensitive to the structural information present in the scene. This method incorporates a new saliency detection method by considering spatial and temporal aspects of the video sequance. The Correlation Coefficient (CC) calculated for the obtained results shows that the values of the derived metric are well correlated with the corresponding subjective test results.