Perceived color quality and its effects on video: Modeling and prediction

In this dissertation, our goal is to quantify the video and color quality of digital videos by focusing on the process of varying compression bit rate and varying colorfulness (lightness, chrome, or hue scaling). These processes were studied to determine how they interact in influencing perceived color quality and perceived video quality. We used four psychophysical experiments to collect subjective scores on three specific attributes from the processed videos: perceived color preference, perceived color naturalness, and perceived overall annoyance.The results of the experiment indicate that subjects may like the colors of processed videos but at the same time consider these colors less than optimally natural. Furthermore, subjects may prefer the colors of compressed video or perceive the colors to be natural but yet consider the combination of compression defects and color scaling annoying. Of the three color scaling methods, chrome scaling resulted in consistent increases in preference and naturalness scores or reductions in annoyance scores from the unprocessed videos. Lightness scaling also offers some increases in preference and naturalness scores or reduction in annoyance from the processed videos. However, these gains are much smaller compared to chroma scaling. Videos containing natural images (non man made objects) are generally unacceptable when its hue is manipulated. However, with videos that contain man made objects, some hue manipulation is acceptable. Thus, global hue manipulation is best avoided.Simple linear, Logistic, or Gaussian functions are effective in describing the effect of either varying compression bit rate (represented as logarithm of the total squared error) or varying color content (represented as mean lightness, chroma, or hue). From these models of single effects, models of combined effects can be built that perform relatively well in some cases. Finally, using statistics extracted from the video's physical signal, including color statistics, spatial statistics, temporal statistics, and artifact strengths, models can be developed that account for the effect of varying compression hit rate, colorfulness, and video content. These models are applicable in benchmarking simple color enhancement (e.g. chroma and lightness scaling), as a tool in an adaptive color enhancement, or to improve existing video quality metrics.