Scalable Lossless High Definition Image Coding on Multicore Platforms

Abstract

With the advent of multicores in all processor segments including mobile, embedded, desktop and server ones, we are in the new era of multiplying computing power via scaling the number of cores. The multicore approach is more versatile and programmable than the ASIC approach. For instance, the same multicore product can be adapted to the ever-improving potpourri image processing standards. Developing ASIC modules for each standard would pose million-dollar start-up cost and time-to-market disadvantage. However, the multicore approach is a two-edge sword: Unleashing its multiplying power presents significant programming challenges. The harmony between the multiplying power and programming productivity is the holy grail in this field. This paper addresses the challenge in the Digital Cinema domain. This paper presents an oblivious parallelization paradigm in both compressing and decompressing images via JPEG2000 on multicore platforms with maximum productivity. This approach dramatically reduces compression and decompression time in performing JPEG2000 lossless encoding and decoding algorithms on high definition images in almost real time without any extra hardware acceleration. By boosting parallelism coverage, the high resolution images could be compressed and decompressed in near real time: 15 images decoded/encoded per second. To the best of our knowledge, we are the first to propose a software-based coding solution using commodity multicores to achieve near real-time performance result for JPEG2000. This cost-effective approach could be applied to digital cinema on devices with multicores.