Acceleration of Discrete Wavelet Transform in JPEG2000 by Embedded Super Parallel Processor

摘要

Discrete wavelet transform (DWT) offers compression benefits on high-resolution images. However, processing-time consumption and floatingpoint arithmetic in DWT are expensive parts for some embedded systems. Jasper, .a software-based implementation of JPEG2000, proposed fixed-point arithmetic in DWT to reduce the complexity of DWT but it still consumes large processing-time. In this paper, we present an implementation of DWT based on single-instruction multiple-data (SIMD) architecture of embedded super parallel processor (MX). It helps to reduce the total number of instructions needed for the transformation such as multiplication and addition. Our implementation vectorized the fixedpoint DWT used in Jasper by using vertical channel operations of MX. As a result, the performance of DWT on MX is around 12.37 to 32.37 times faster than conventional implementations on other embedded systems.