Browse Theses

Video streaming service is an essential component of the Internet ecosystem since much of Internet content is consumed via video streaming services. However, the network capacity is a growing bottleneck for video streaming services to deliver high-quality video to end-users. To this end, we consider various and affluent data available in the video streaming ecosystem, which opens lots of opportunities that we can leverage machine learning to reduce the bandwidth usage. In this dissertation, we present machine learning based approaches to reduce the bandwidth usage in on-demand video streaming.In the first part of this dissertation, we present SALI360 that utilizes the characteristics of the human vision system (HVS) to solve the bandwidth requirement. By pre-rendering a set of regions where viewers are expected to fixate on VR content in higher quality than the other regions, SALI360 improves viewers' quality of perception (QoP) while reducing content size with geometry-based 360-degree content encoding. Results of our experimental VR video streaming show that SALI360 achieves 53.3% of PSNR improvement in viewers' perception area. In addition, our subjective study on 93 participants verifies that SALI360 improves viewers' QoP in the 360-degree VR streaming service.In the second part of this dissertation, we present dcSR, a lightweight video super resolution (SR) approach that enables a practical neural quality enhancement. dcSR takes a data-centric AI paradigm that targets improving the data consistency for training. On the server-side, dcSR constructs micro SR models trained on a few selected frames from each video by employing a long-term video scene understanding mechanism. On the client-side, dcSR integrates the micro-SR models into the video decoder and enhances the video quality in real-time without compromising quality enhancement. Weevaluate dcSR and show its benefits by comparing it with previous methods.Finally, we present SenseQ, a context-aware mobile video quality adaptation method that adjusts the video quality, by analyzing various data accessible in the mobile video streaming system. By understanding the complex relationship between the context data and users' quality perception, SenseQcan reduce the network usage in the video streaming system, while maintaining the comparable quality perception. In the evaluation, we demonstrate that SenseQ shows a potential for network saving, maintaining the comparable quality rating. In addition, we show that the prediction capability ofSenseQ enables us to utilize SenseQ to estimate the network usage and users' expected quality rating over the different network scenarios.