A parallel implementation of the 2-D discrete wavelet transform without interprocessor communications

The discrete wavelet transform is currently attracting much interest among researchers and practitioners as a powerful tool for a wide variety of digital signal and imaging processing applications. This article presents an efficient approach to compute the two-dimensional (2-D) discrete wavelet transform in standard form on parallel general-purpose computers. This approach does not require transposition of intermediate results and avoids interprocessor communication. Since it is based on matrix-vector multiplication, our technique does not introduce any restriction on the size of the input data or on the transform parameters. Complete use of the available processor parallelism, modularity, and scalability are achieved. Theoretical and experimental evaluations and comparisons are given with respect to traditional parallelization.