Design of parallel multiresolution filter banks by simulated annealing

A new simulated annealing technique is described for designing multiresolution, parallel- structured filter banks with finite wordlength coefficients. The algorithm is based on a two stage optimization. In the first stage, a perturbation following uniform distribution is used; while in the second stage, perturbation is in Gaussian distribution, allowing a faster convergence. The filter banks have properties which make them well suited to image subband coding applications. In particular, they provide joint localization in the spatial and spectral domains, and allow no leakage of DC into higher subbands. This paper describes the algorithm and presents design examples.