Efficient Distributed Coding of Correlated Images

Dr. David W. Pan

Department of Electrical & Computer Engineering
UA Huntsville

October 17, 2008
219 Shelby Center
3:00 (Refreshements at 2:30)

Abstract

In many image compression applications such as visual sensor networks, there is usually an asymmetry in resources between the encoder and decoder, where the encoder has sparse computational resources and compresses an image by exploiting its statistics. The compressed image is transmitted over some channel, and a decoder with abundant computational resources decompresses and reconstructs the image. Such applications can benefit from the use of distributed source coding techniques, which achieve image compression by exploiting source statistics mainly at the decoders, and thus have the distinct advantage of allowing for low-complexity encoders desired in many resource-constrained applications. The compression efficiencies of these emerging distributed source coding methods depend heavily on how well the source correlations can be captured and exploited. We will give an overview of the existing distributed source coding techniques and discuss some recent advances in modeling geometric correlations of images in low dimensional spaces. Research in this area is of multidisciplinary nature, exploring the common ground of electrical engineering, digital communication, mathematics, and computer science.