A FLEXIBLE ERROR RESILIENT SCHEME FOR JPEG 2000 Tammam Tillo, Marco Grangetto, Gabriella Olmo CERCOM - Center for Multimedia Radio Communications Dipartimento di Elettronica - Politecnico di Torino Corso Duca degli Abruzzi 24 - 10129 Torino - Italy Ph.: +39-011-2276510 - FAX: +39-011-5644099 tammam.tillo@polito.it, marco.grangetto@polito.it, gabriella.olmo@polito.it ABSTRACT Nowadays, wireless multimedia applications are experienc- ing a rapid growth; in this scenario, challenging obstacles, such as packet losses due to congestion and band limitation along with bit-level error corruption, require the design of novel solutions for robust multimedia delivery. New stan- dards for multimedia applications are incorporating many tools for error resilience; as an example, JPEG 2000 part 11 is explicitly devoted to the wireless applications of the im- age coding standard. In this paper we address a novel multi- ple description coding technique, based on post-processing rate allocation of embedded bitstreams. The proposed ap- proach is complaint with the JPEG 2000 standard and has the ability to jointly cope with both packet losses and bit errors. Experimental results show that the designed algo- rithm significantly outperforms other techniques based on unequal error protection by means of RS codes. 1. INTRODUCTION It is well known that multimedia applications generate a huge amount of data, which require a large bandwidth in or- der to be transmitted. Therefore, the use of efficient source coding, able to represent the data in the most possible com- pact form, is mandatory. Generally, the coded stream is highly vulnerable to transmission errors because most of the source redundancy has been removed. Moreover, in the case of wireless real time multimedia transmission, the UDP protocol with only IP checksum is typically used at transport layer [1]; as a consequence, not only the network can introduce packet loss but also the packet payload can be delivered with residual bit errors to the application layer. For this reason, error resilience techniques are widely stud- ied and many tools are proposed in recent standards, such This work was partially supported by the Italian Ministry of Education and Research under the CERCOM (Center for Multimedia Radio Commu- nications) and PRIMO (Reconfigurable Platforms for Wideband Wireless Communications) grants. as JPEG 2000 [5]. These tools yield a reasonable level of robustness, but they may fail to guarantee acceptable qual- ity in harsh transmission contexts such as the wireless one. Consequently, the use of forward error correction codes, in order to protect the bitstream, is common practice. In [2], turbo codes are used to protect the JPEG 2000 bitstream. In [3], Reed Solomon codes are applied in order to provide unequal error protection UEP of the JPEG 2000 bitstream. In [4], a JPEG 2000 backward compatible header error pro- tection mechanism is described. In this paper, we propose a novel method to jointly pro- tect the JPEG 2000 codestream against packet loss and resid- ual bit errors. Simulation results demonstrate that the pro- posed scheme exhibits a significant gain when compared to the UEP strategy in [3]. Besides the error resilience per- formance, the main features of the technique are its limited computational complexity at both the encoder and decoder side, and its full compatibility with a standard JPEG 2000 decoder. Moreover, the proposed scheme can be flexibly configured according to the application requirements in or- der to select the best trade-off between packet loss robust- ness, and bit level error protection. 2. GENERATING JPEG 2000 RESILIENT STREAMS In JPEG 2000, the various components of the tiled image are transformed by means of the discrete wavelet transform. The transformed coefficients are grouped in codeblocks (CB), which are independently coded one bit-plane at a time; then, a post-processing rate allocation step determines the extent to which each CB bitstream should be truncated. This ap- proach assures that a CB bitstream encoded at low bit rate is completely embedded in a higher rate one. As a conse- quence, the bitstream of an image encoded at rate R 2 <R 1 is entirely included in that encoded at rate R 1 . Alternatively, two bitstreams at rates R 1 and R 2 <R 1 can be considered as two different representations of the original image; the