PhD Defense: "Advances in Video Coding Based on Principles of Optimal Estimation"

Bohan Li

June 11th (Tuesday), 4:00pm
Engineering Science Building (ESB), Rm 2001

Recently, the amount of video contents has seen a dramatic growth and video coding applications such as video streaming, sharing and real-time conferencing are playing a more and more important role. The increasing demand of high-quality video contents and limitations in network bandwidth and storage present great challenges to the video compression field.

This talk investigates one of the crucial components of video compression, predictive coding, based on optimal estimation principles, and is composed of two parts. First, the prediction scheme of error-resilient video coding with lossy networks is investigated, where the accurate estimation of end-to-end distortion (EED) is crucial for the encoder to perform optimal decisions. As a first step, existing EED estimation approaches are extended to account for recent advanced coding tools. Furthermore, it is recognized that the existing tools are not designed for lossy networks, and thus a novel framework specifically tailored for this situation is proposed with a soft-reset prediction mode, where the encoder is able to fine-tune the error propagation and thus achieves a significant performance gain. As another example, we also establish EED estimation recursions for state estimation of wireless sensor networks and proposed an adaptive approach to account for channel errors with Kalman filters, which further proves the significance of EED estimation.

The second part shifts focus to the bi-directional motion compensated prediction scheme in video coding. To overcome the issue in the conventional scheme that the motion information at the decoder is not efficiently utilized, a novel framework with the co-located reference frame (CLRF) is proposed, where a reference frame is interpolated by the motion field estimated between the references at the decoder, and an extra step of block-based motion estimation is then performed to correct possible motion offsets. Moreover, since estimating motion field at the decoder suffers from quantization error and significant complexity rise, we then propose to apply an estimation-theory based approach to utilize the motion vectors (MVs) already available to the decoder, where an optimal linear estimator is then adaptively derived and used to interpolate the CLRF. The available MV candidates are then also utilized to predict the MV of the current block, such that the prediction is most consistent with the observations given a certain pixel correlation model. Experimental results show that the proposed approach provides significant coding performance improvement while maintaining a reasonable complexity.

About Bohan Li:

Bohan Li received his B.S. degree in Electronic Engineering from Tsinghua University in 2014, and his M.S. degree in Electrical and Computer Engineering from University of California, Santa Barbara in 2016. He is currently working towards his Ph.D. degree at the Signal Compression Lab (SCL) in ECE department, UCSB. His research focuses on error resilient coding and estimation theoretic predictive coding for video compression.

Hosted by: Professor Kenneth Rose