PARALLEL PIPELINE IMPLEMENTATION OF WAVELET TRANSFORMS

Wavelet transforms have been one of the important signal processing de velopments in the last decade, especially for applications such as tim e-frequency analysis, data compression, segmentation and vision. Altho ugh several efficient implementations of wavelet transforms have been derived, their computational burden is still considerable. The paper d escribes two generic parallel implementations of wavelet transforms, b ased on the pipeline processor farming methodology, which have the pot ential to achieve real-time performance. Results show that the paralle l implementation of the oversampled wavelet transform achieves virtual ly linear speedup, while the parallel implementation of the discrete w avelet transform (DWT) also outperforms the sequential version, provid ed that the filter order is large. The DWT parallelisation performance improves with increasing data length and filter order, while the freq uency-domain implementation performance is independent of wavelet filt er order. Parallel pipeline implementations are currently suitable for processing multidimensional images with data length at least 512 pixe ls.