A New Approach to Implement Discrete Wavelet Transform on Coarse-Grained Reconfigurable Architecture

Abstract

Discrete Wavelet Transform (DWT) is widely-used in image and video processing with high computing complexity and regular data flow, which is suitable for the implementation on a Coarse-grained Reconfigurable Architecture (CGRA) owing to its rich parallel computing resources. In this article, the two wavelet filters adopted in JPEG2000 image standard, 5/3 DWT and 9/7 DWT, were realized on a CGRA platform called Reconfigurable Multimedia System-II (REMUS-II). The result shows that the CGRA-based implementation has advantage in area, power and performance over the state-of the-art GPU including 7800GTX and 9800GTX. The die size and power consumption of REMUS-II is respectively less than 1% and 10% compared to the GPU implementations, whereas the performance speed-up is 92.9x for 9/7 filter compared to GPU 7800GTX and 6.54x for 5/3 filter compared to GPU 9800GTX.