MSc Semester Project
Title
Multiview Distributed Video Coding
Candidate
Nicolae Cleju
Supervisor
Prof. Touradj Ebrahimi
Assistant: Mourad Ouaret
Expected
2008, summer semester
Place
EPFL
Description
Distributed Video Coding (DVC) is a new concept in the field of video coding. Traditionally, the burden of compression has been on the encoder side. Namely, the encoder analyzes the content in order to represent it in the most compact and cost efficient way, leading to a system with complex encoders and simple decoders especially suited for broadcasting applications. Distributed Video Coding is based on a completely different paradigm. More specifically, a simple encoder generates multiple correlated streams and a complex decoder exploits the redundancies between these multiple streams. This new paradigm is especially appealing in mobile applications.
In this project, we propose to develop a multiview video coding system based on DVC and to evaluate its performance. The idea is to fuse different side informations to construct a final one that will be used in the DVC decoding process. It is obvious that the fusion should be done in a way that the compression efficiency is improved.
The student will have to perform the following tasks:
- Study Distributed Video Coding and state of the art of view prediction techniques, especially ones used in multiview.
- Perform fusion between MCTI and MVME side informations using a partially decoded frame, obtained using some initial amount of Wyner-Ziv bits.
- Implement the fusion tool into the DVC multiview software and evaluate its performance.
- Deeply analyze when and under which conditions MVME wins against MCTI.
- The sequences Breakdancers, Ballet and Uli should be used in the evaluation.