UC San Diego

With multimedia research burgeoning, video applications have become essential to our daily life. However, as the compression becomes more aggressive, too much data loss results in degrading perceived video quality for viewers. Therefore, an accurate quality measurement is important to improve or preserve the quality of compressed video. This dissertation focuses on measuring the quality degradations that are caused by compression. We specifically target distortions with impact above the human perceptual threshold, which are also called artifacts. This type of distortion usually appears in a structured form. This characteristic makes quality assessment highly content dependent and many existing metrics fail in this regard. Some previous research has tried to raise the accuracy of video quality assessment by considering human visual system (HVS) effects, or human visual attention factors. However, both HVS and human visual attention have very strong interaction in the video quality assessment process, and none of the existing quality measurement research takes both of them into account. In addition, cognitive factors significantly influence the visual quality assessment process, but they have been ignored in current quality assessment research. Based on these realizations, a new video quality assessment philosophy is introduced in this thesis. It considers the characteristics of artifacts, effects from HVS, visual attention, and cognitive non- linearity. First, a new human visual module is proposed, it takes both visual masking and attention effects into account. Its unique design makes embedding this visual module in any video quality related applications very easy. Based on this new human visual module, a blurriness metric is designed which includes cognitive characteristics. This new blurriness metric does not rely on edge information, and is more robust at assessing heavily compressed video data. A metric for artifacts introduced by motion compensated field interpolation (MCFI) is also implemented. It is the first metric ever designed for measuring the spatial quality of temporally interpolated frames. From a temporal quality perspective, a novel temporal quality metric is designed to measure the temporal quality degradation caused by both uniform and non-uniform distributed frame loss. Experimental data shows these metrics significantly outperform the existing metrics