Scalable DSP realization of wavelet transform in image coding

A scalable realisation of a two dimensional (2D) fast wavelet transform (FWT) is presented and compared to an earlier implementation of discrete wavelet transform (DWT), which uses matrix multiplication method. Parallelisation and mapping possibilities are analysed. The main emphasis is in minimising communication requirements and utilising local communication. Measured performance figures verify good performance and illustrate the benefits of parallel computation in Wavelet transform. The computation time of the FWT is only a fraction of the computation time of the previous implementation.