PERCEPTUAL QUALITY ASSESSMENT OF VIDEOS AFFECTED BY PACKET LOSSES by

Due to rapid advance of various video applications and services, such as video telephony, mobile video broadcasting, and Internet Protocol television (IPTV), there is an increasing demand for accurate and effective quality assessment of underlying videos. Accurate video quality assessment is crucial to video codec development, network protocol planning, in-network quality monitoring, quality assurance of end users, etc. This thesis develops several objective quality metrics for videos impaired by packet losses. Because transmission error is one of the main causes of quality degradations of networked video, a deep investigation on impacts of different attributes of packet losses on perceptual video quality is performed. Based on the observed relationships between perceptual video quality and various attributes of packet losses, e.g. error length, the loss severity, loss location, the number of losses, and loss patterns, we incorporate a prior quality metric for coding artifacts and propose a novel video quality metric considering both coding and packet-loss artifacts. In the hope of improving the accuracy of quality metric for video sequences, we perform another study which focuses on quality assessment of single packet-loss-affected video frames. We evaluate the impacts of various properties of human visual system on quality of video frames, and develop quality metrics considering coding and packet-loss artifact, first separately and then jointly. In order to further improve the prediction performance of existing quality metrics, we exploit several methods of incorporating saliency into a video quality metric. The better performance of proposed saliency-aided quality metrics confirms the significant role of saliency in video quality assessment. Finally, we extend our study on saliency-aided video quality assessment to prediction of packet loss visibility. We update an existing loss visibility predictor with saliency-based features, and show that considering saliency can lead to improved prediction accuracy. To explore relationships between various video attributes and perceptual video quality, we carefully design and perform extensive subjective video quality tests. The obtained subjective results not only confirm our assumptions about such relationships, but also inspire us to pursue our research in several novel directions. These subjective data also show fairly high correlations with the proposed objective quality measures. Figure 4.8 Reference frames from the sequence " aircraft " and " leaf " ; Distorted frames from the sequence " aircraft " and " leaf " .. Figure 4.10 SM 1 (t) and SM 2 (t) of " optis " , " aircraft " , " leaf " ; …