Discrete wavelet transform on consumer-level graphics processing unit

The discrete wavelet transform (DWT) has been extensively studied and developed in various scientific and engineering fields. The multiresolution and local nature of the DWT facilitates applications requiring progressiveness and the capture of high-frequency details. However, the intensive computation of DWT caused by multilevel filtering/down-sampling will become a significant bottleneck in real-time applications when the data size is large. This paper presents a SIMD-based parallel processing framework as a commodity solution to this problem, that is based on the consumer-level programmable graphic processing unit (GPU) on personal computers. Simulation tests show that, in contrast to those CPU-based solutions for DWT, this GPU-based parallel processing framework can bring a significant performance gain on a normal PC without extra cost.