Imperial College

Research Group

Distributed Source Coding

In conventional source coding, a single encoder exploits the correlation of the source to perform compression. Many applications, however, involve multiple sources often separated in space that need to be compressed independently. Distributed source coding therefore studies the problem of compressing such sources in a distributed fashion. Distributed source coding finds application in sensor networks, low-complexity video encoding, compression of multi-view images and multi-view videos.

Our main interest is in developing new distributed compression algorithms for compression of multi-view images or video sequences. We are exploring the use of quad-tree decomposition algorithms and extensions of the wavelet transform to develop new distributed image or video compression schemes.  Our algorithms are tested using the multicamera system shown in Fig 1 and sponsored by Royal Society and Selex.

Camera System
Fig. 1. Our multiview camera system sponsored by the Royal Society and Selex.

To achieve distributed compression of multi-view images, we developed a quad-tree based compression algorithm. In our approach some images are compressed using this algorithm and the full information is transmitted, for other images only partial information of the quadtree structure and of the leaf content is transmitted. The receiver estimates the missing information by decoding the information received jointly.

Simulation Results

Distributed Compression

Distributed Compression


Main Publications:

PhD Student: Nicolas Gehrig, Varit Chaisinthop.

Collaborations and Interactions:  M. Vetterli (EPFL) and  K.  Ramchandran  (UC, Berkeley) 

Department Home