Reduction of the memory requirements for the VLSI implementation of the 2D-inverse fast wavelet transform, using a space-filling curve

Since wavelet based codecs seem to be a promising alternative to the classically used block-transform based codecs, it is important to take a look at the hardware implementation constraints that are imposed by such multiresolution architectures. We show that by reorganising the data order within the 2D-FWT and taking the causality relation between the 2D-FWT and 2D-IFWT into account, the memory requirements and latencies for practical 2D-FWT/2D-IFWT compression/decompression schemes are minimised by using specific 2D fractal space-filling image scan curves