Perceptual video quality and quality of experience for adaptive video streaming

Abstract

We live in a world where images and videos dominate our everyday lives. Every day, an enormous amount of video data is being shared in social media and consumer applications, while video streaming is becoming a new form of digital entertainment. Large-scale video streaming on demand has become possible thanks to numerous engineering achievements in fields such as video compression, high-speed computation and display technologies. Nevertheless, the skyrocketing needs for bandwidth and network resources consumed by video applications challenges modern video content delivery. Since the available bandwidth resources are limited, streaming service providers have to mediate between operation costs, bandwidth efficiency and maximizing user quality of experience. However, these goals are inherently conflicting and require knowledge of how user quality of experience is affected by the network-induced changes in video quality. Being able to understand and predict user quality of experience and perceptually optimize rate allocation, can have significant effects in better network utilization, reduced costs for service providers and improved user satisfaction. The goal of this dissertation is to study and predict user quality of experience in video streaming applications, by exploiting perceptual video quality and human behavioral responses to streaming-related video impairments. To this end, I present the details of three large-scale video subjective studies which target video streaming under multiple viewing conditions, such as display device, session duration, content characteristics and network/buffer conditions. By analyzing how humans react to changes in visual quality and streaming video impairments, I also design numerous video quality and quality of experience prediction models that can be used to evaluate the overall and the continuous-time perceived video quality. Throughout this dissertation, my goal is to perceptually optimize various stages of the video streaming pipeline, such as video encoding and video quality control as well as client-based rate adaptation. Ultimately, I envision that the outcome of this dissertation can be useful for video streaming applications at global scale