Optimized GPU implementation of JPEG 2000 for satellite image decompression

Abstract

JPEG 2000 is a powerful yet computationally complex image compression algorithm which is widely used in remote sensing applications. In this work, we focus on speeding-up the decompression algorithm by applying various GPU optimization techniques. We have conducted numerous experiments on high-resolution satellite images in two operational modes; a syn-chronous mode and an asynchronous batch mode. We share our experiment results and make performance evaluations regarding each operational mode and optimization method separately. Finally we propose an optimized GPU architecture for satellite image decompression and compare the achieved performance with a multi-threaded CPU architecture. © 2018 IEEE.