Abstract:

Since the classical wavelet transform algorithm on CPU hardly meets the realtime performance requirements, especially dealing with large scale data in high resolution, we presented a GPUbased parallel wavelet transform algorithm, which improves the locality of local memory access and increases the bandwidth of global memory access. It uses multikernel to improve the performance in the case of multiresolutions. The experiment results show that compared to the performance of a classical algorithm on CPU, GPU gains the speedup of 30—60, accordingly, it can satisfy the realtime requirements for transformation.

Key words: wavelet transform, multiresolutions, GPU acclerate