PhD Defense: "Toward High Precision Predictive Coding in Video Compression"

Wei-Ting Lin

July 25th (Thursday), 11:00am
Harold Frank Hall (HFH), Rm 4164 (ECE Conf Rm)

Nowadays, video streaming and content delivery services have become an essential part of our lives. As the demands for fast and ultrahigh-quality videos continue to grow, more and more loads will be placed on our limited network and storage systems. The development of advanced video compression algorithm is crucial to relieve the burden on the systems and meet user requirements.

In this talk, I will focus on one of the crucial components in video compression, predictive coding, wherein the temporal correlations within a video sequence are utilized to reduce data redundancy. First, the sub-optimality of the conventional block-based motion compensation (BMC) method, which is widely used in all recent coding standard, is investigated. To address the shortcomings of BMC, a novel framework of adaptive interpolated motion compensation (AIMC) is proposed based on the optimal estimation theory, where the motion vectors (MVs) are explicitly treated as pointers to sources of observations. I will first present a non-parametric framework wherein the corresponding optimal linear estimation coefficients for combining the relevant observations to form the final prediction to a pixel are directly derived from training data. Prediction coefficients are further adapted to local statistics by switching between predefined sets of coefficients, which are trained offline through a procedure of “K-modes” clustering. As the additional side information for indicating the set of prediction coefficients used to generate the final prediction is generally not negligible, a parametric framework is proposed to model the statistics of observations and the target pixel. The coefficients derived from the model can automatically adapt to local variations without additional side information. The experimental results show significant coding performance improvement while maintaining a reasonable complexity.

About Wei-Ting Lin:

photo of Wei-Ting Lin Wei-Ting Lin is a PhD candidate in the Department of Electrical and Computer Engineering at University of California, Santa Barbara. He received his M.S. degree in Graduate Institute of Communication Engineering from National Taiwan University in 2012, and his B.S. degree in Electrical Engineering from National Taiwan University in 2010. His research focuses on estimation theoretic predictive coding and system design for video compression.

Hosted by: Professor Kenneth Rose