Multiple description image coding framework for EBCOT

Koduvayur P. Subbalakshmi, Siva Somasundaram

Abstract: Multiple description coding (MDC) schemes have become popular as robustness tools in packet switched networks where multiple paths with varying packet loss rates are available. The descriptions are designed so that the quality of the reconstructed image depends on the number of descriptions received and not on which descriptions are actually received. We propose a new MDC framework for the EBCOT algorithm used in the current JPEG 2000 standard for still image compression. The inherent block coding in EBCOT allows for more flexibility in the formation of basic groups that are then combined in different ways to form descriptions. The bit allocation problem (between the different groups in a description) is formulated and optimized for good rate-distortion performance. Optimization of the bit allocation between the primary and secondary positions in each description is obtained using the generalized BFOS algorithm. Experiments using the Kakadu software (for JPEG 2000) show that the performance of the proposed framework degrades gracefully with increase in the number of descriptions lost.