【作者】 战晓明；

【导师】 徐炜遐； 窦勇；

【作者基本信息】 国防科学技术大学 ， 计算机科学与技术， 2006， 硕士

【摘要】 JPEG2000是联合图片专家组(Joint Photographic Experts Group, JPEG)于2000年制定的一个静止图像压缩编码国际标准,与JPEG相比,它具有压缩率高、抗干扰性强、可选择有损和无损压缩、多分辨率表示、感兴趣区域访问等特点。但是,目前JPEG2000尚不成熟。首先,其标准并未完成,还在不断补充之中;其次,由于JPEG2000的核心编解码器采用小波变换、算术编码及嵌入式分层组织,较以往的静止图像压缩标准复杂度增加很多,目前的实现技术很难做到与JPEG同等的处理速度。为了解决JPEG2000系统实现中的问题,本文对JPEG2000标准进行了深入分析,研究了其主要算法,找出了性能瓶颈,提出了软件和硬件两个方面的优化方法,并予以实现。软件方面,首先利用Intel Pentium 4平台的SIMD扩展指令,研究了多分量颜色变换、离散小波变换的多种加速方法。其次,通过对该平台存储子系统的研究,提出了针对离散小波变换的分块和预取优化方法。对于难以通过软件获得性能提升的算术编码部分,设计并实现了基于VLSI的硬件结构,达到了加速的目的。本文对于每种优化方法都给出了性能分析和实验结果。更多还原

【Abstract】 JPEG2000 is a still image compression standard presented by Joint Photographic Experts Group in 2000. It provides many advantages over the previous JPEG, such as high compression efficiency, error resilience, lossy and lossless compression, multiple resolution representation, region-of-interest (ROI). But today JPEG2000 has not gain its success. The standard is not completely done and is still being complemented. Besides, the core codec implicates discrete wavelet transform, arithmetic coder and embedded bit-stream, which are more complicated than the previous standards. It is unlikely to achieve the same speed with JPEG based on today’s implementations.Provided the problems above, this paper makes a research of the JPEG2000 standard and its encoding flow. Both software and hardware optimizations are made.Software works are based upon Intel Pentium 4 platform. By using SIMD extensions, multi-component transform and discrete wavelet transform are accelerated. By researching the memory and cache subsystem, blocking and prefetching methods are proposed in the target of enhancing the proformance of discrete wavelet transform.For the arithmetic coder which is hard to gain speedup by software, a VLSI design is presented, and comes to the purpose of acceleration. For each method in this paper, performance result is analyzed.更多还原