Multiple Description Image Coding: A New Efficient and Low Complexity Approach for Wireless Applications Mehdi Malboubi, Ahmad Bahai, and Mustafa Ergen Department of EECS University of California Berkeley mehdi_malboubi@yahoo.com, {bahai, ergen}@eecs.berkeley.edu Abstract This paper describes a new efficient and simple multiple description image coding system for the reliable communication of images in wireless communication systems. This system uses the inherent redundancy in an image to create several balanced descriptions. Each description is coded using a minimum complexity JPEG image coder and at the decoder, based on the number of received descriptions, an image with different qualities is reconstructed. The simulation results indicate the good performance of this method, particularly for wireless applications. 1- Introduction A communication system is designed to transmit information reliably. In wireless communication systems, due to packet loss and channel failures, reliable communication is a proven challenge. Multiple Description (MD) coding is an effective method to create diversity and consequently enhance the reliability of communication systems to meet the challenge. In this method, several independent and equally important descriptions of a source are produced and simultaneously transmitted over several channels, so that various reconstruction qualities are obtained from different subsets of descriptions [1, 2, 3, 4]. For this purpose, different methods have been introduced that can be divided into two important classes, Bandwidth-Efficient and Non-Bandwidth-Efficient methods [5]. Bandwidth-efficient MD encoders use inherent redundancy in the signal and do not introduce extra redundancy between descriptions; in fact, the inherent redundancy makes the bitstream robust and enables lost descriptions to be estimated from received descriptions. In these methods, different descriptions are created by partitioning the domain of the signal, before or after a transformation [5]. While, non-bandwidth-efficient MD encoders introduce a controlled extra redundancy (redundancy which is not inherent in the signal, but added to it) between descriptions in different ways: through overlapping side quantizers (MDSQ) [6], correlating transforms [1], frame expansions [3], or forward error correction (FEC) [7]. For this reason, non-bandwidth-efficient MD encoders typically need more bandwidth and computational power than the bandwidth-efficient MD encoders. Therefore, due to the high