Predicting 3D Quality based on Content Analysis

Development of objective quality metrics that can reliably predict perceived quality of 3D video sequences is challenging. Various 3D objective metrics have been proposed, but PSNR is still widely used. Several studies have shown that PSNR is strongly content dependent, but the exact relationship between PSNR values and perceived quality has not been established yet. In this paper, we propose a model to predict the relationship between PSNR values and perceived quality of stereoscopic video sequences based on content analysis. The model was trained and evaluated on a dataset of stereoscopic video sequences with associated ground truth MOS. Results showed that the proposed model achieved high correlation with perceived quality and was quite robust across contents when the training set contained various contents.