Parallel architectural design space exploration for real-time image compression

Abstract

Embedded block coding with optimized truncation (EBCOT) is a coding algorithm used in JPEG2000. EBCOT operates on the wavelet transformed data to generate highly scalable compressed bit stream. Sub-band samples obtained from wavelet transform are partitioned into smaller blocks called code-blocks. EBCOT encoding is done on blocks to avoid error propagation through the bands and to increase robustness. Block wise encoding provides flexibility for parallel hardware implementation of EBCOT. The encoding process in JPEG2000 is divided into two phases: Tier 1 coding (Entropy encoding) and Tier 2 coding (Tag tree coding). This thesis deals with design space exploration and implementation of parallel hardware architecture of Tier 1 encoder used in JPEG2000. Parallel capabilities of Tier-1 encoder is the motivation for exploration of high performance real time image compression architecture in hardware. The design space covers the following investigations: - The effect of block-size in terms of resources, speed, and compression performance, - Computational performance. The key computational performance parameters targeted by the architecture are - significant speedup compared to a sequential implementation, - minimum processing latency and, - minimum logic resource utilization. The proposed architecture is developed for an embedded application system, coded in VHDL and synthesized for implementation on Xilinx FPGA system.