GPU-Based DWT Acceleration for JPEG2000
Jiří Matela · 2009
Abstract. In our paper, we focus on accelerating DWT (Discrete Wavelet Transform) part of the JPEG2000 using general-purpose processing on graphical processing unit (GPU). We utilize Compute Unified Device Architecture (CUDA) platform which has its specific properties and constraints. In particular, a proper memory treatment can greatly affect overall performance. In this paper, we briefly describe corresponding CUDA technical background followed by elaborated description of the algorithm—especially its memory layout part. Resulting implementation of DWT performs very well compared to other implementations available and is able to process an HD sized picture within 1 ms or faster. 1