Distributed Compression in Camera Sensor Networks
Abstract
This report results from a contract tasking Imperial College London as follows: This effort will address a distributed compression problem using information theoretic methods originating in the work of Slepian and Wolf for lossless compression and extended by Wyner and Ziv to the case of lossy compression of continuous-valued sources. The theories developed in these papers are non-constructive and rely on asymptotic random coding arguments. Constructive designs of encoders for the distributed compression problem based on channel codes have been subsequently proposed with applicability to sensor networks. However, in a realistic context the statistics of the source are not known a-priori and channels codes such as turbo or trellis codes might be too complicated in this context. This effort will make use of the correlation structure of the data given by the plenoptic function in the case of multi-camera systems. In many cases the structure of the plenoptic function can be estimated without requiring inter-sensor communications, but by using some a-priori global geometrical information. Once the structure of the plenoptic function has been predicted, it is possible to develop specific distributed compression algorithms that do not require the use of complicated channel codes. This effort will develop techniques to predict the structure of the plenoptic function and develop very simple and efficient distributed compression algorithms derived from a design of a new fully distributed image compression scheme for multi-view images. The algorithm will be implemented in Matlab or C and will operate on some sets of pre-selected multi-view images. Commented MatLab psuedo-code and or C code will be provided with any executables demonstrating the algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 13, 2006
- Accession Number
- ADA454468
Entities
People
- Pier L. Dragotti
Organizations
- Imperial College London