Pipelined recursive filter architectures for subband image coding

Abstract

Recursive-filter-based structures have been recently put forward as a potentially beneficial alternative to the more conventional two-band FIR QMF filter banks for applications in image subband analysis/synthesis. Zero-phase causal/anticausal IIR all-pass filter pairs make it possible to achieve perfect reconstruction, with a markedly better computational efficiency than FIR-based solutions. A major a priori hindrance, namely the iteration bound that sets an upper limit on the throughput achievable by pipelining in any implementation of recursive filters, has not been specifically addressed in this application context so far. We study the application of 2-D interleaving and filter transformation techniques to the pipelining of the polyphase subband tree. These transformations are combined with relevant arithmetic strategies for a comparative assessment of VLSI architectural implementations of image subband coding-decoding systems, targeting high-throughput real-time applications in visual communication.