Distributed video coding (DVC) is a compression paradigm based on two key Information Theory results from the 70s: the Slepian-Wolf and the Wyner-Ziv theorems. Contrary to the available predictive video codecs, DVC codecs exploit the source statistics at the decoder providing functional benefits such as: 1) flexible distribution of the computational complexity between encoder and decoder; 2) in-built robustness to channel losses; 3) exploitation of Inter-view data correlation at the decoder.
These functional benefits are very important for visual sensor networks, a network of wireless low-power video camera nodes, used for a wide range of applications such as surveillance, monitoring of remote areas or tele-presence systems. Visual sensor networks are characterized by: 1) Inter-view correlation between camera nodes with overlapped views; 2) up-link communication between many encoders (one at each camera node) to a few decoders; 3) independent compression of each view, since collaboration between cameras is difficult to implement or rather inefficient.
Due to these characteristics, it is not efficient (or possible) to compress each captured view at a central node with the typical predictive approach, e.g. using the recent multi-view video coding H.264/MVC (Multi-view Video Coding) standard, since it implies to exploit spatial, temporal and Inter-view correlations at a rather high complexity encoder. The distributed coding approach provides a significant architectural benefit since it does not require any Inter-camera communication to explore the Inter-view correlation while the predictive approach requires the availability of all views at some central encoding location, forcing communication between cameras. This project will advance state-of-the-art by developing competitive multi-view DVC solutions for visual sensor networks, notably by proposing:
1. Side information (SI) creation and correlation noised modeling: To fully exploit the potential of the multi-view distributed video coding paradigm, high quality side information (an estimation of the source created at the decoder) must be created by (jointly) exploiting temporal and Inter-view correlations. In this case, solutions that combine temporal and Inter-view SI estimators by using preliminary decoded data as helper information will be proposed. In addition, correlation noise models capturing the statistical correlation between the side information (available at the decoder) and the source (available at the
encoder) will also be proposed along with the estimation of the relevant parameters.
2. Hybrid predictive-distributed video coding: The combination of predictive coding techniques, to exploit the temporal correlation, with distributed coding techniques, to exploit the Inter-view correlation, may be a successful “cocktail” to obtain improved rate-distortion (RD) performance. This approach was never exploited before (as much as the authors know) and will require to reinvent popular coding techniques at both encoder and decoder. In this case, it is also expected to obtain error resilience improvements, which is quite important in this type of error-prone networks (typically wireless).
|Start Date: 01-03-2012|
|End Date: 01-06-2015|
|Team: João Miguel Duarte Ascenso, Catarina Isabel Carvalheiro Brites Ascenso, Fernando Manuel Bernardo Pereira, Dhiraj Kumar Shah|
|Groups: Multimedia Signal Processing – Lx|
|Local Coordinator: João Miguel Duarte Ascenso|