Quality assessment of a stereo pair formed from decoded and synthesized views using objective metrics

When a stereo pair is formed from a decoded view and a synthesized view, it is unclear how the overall quality of the stereo pair should be assessed through objective quality metrics. In this paper, this problem is addressed considering a 3D video represented in the format of multiview video plus depth. The performance of different state-of-the-art 2D quality metrics is analyzed in terms of correlation with subjective perception of video quality. A set of subjective data collected through formal subjective evaluation tests is used as benchmark. Results show that the measured quality of the decoded view has the highest correlation with perceived quality. If the objective quality assessment is based on the measured quality of the synthesized view, it is suggested to use VIF, VQM, MS-SSIM, or SSIM since they significantly outperform other objective metrics, including PSNR.