JPEG2000中高性能低存储的小波变换结构

本文提出了一种JPEG2000中高性能低存储的小波变换结构。该结构通过扩展原始图像数据的精度和有效保护提升步骤中的系数尾数，改善了整型小波变换的性能，提高了图像压缩的质量。同时，为了克服硬件实现中的大量小波系数存储，提出一种码块条带的小波系数存储方案。对子带内的码块条带存储器进行重复利用和有效调度，从存储和功耗两方面减少了硬件资源。实验结果表明，对分辨率为512×512的图像进行小波分解，码块大小选为32×32，采用本文结构的小波系数存储与存储整图小波系数相比可减少80％。整个系统已通过FPGA验证，且综合时钟频率可达到150MHz。

High-performance Memory-efficient Wavelet Transform and Its VLSI Architecture for JPEG2000

This paper proposes a discrete wavelet transform (DWT) architecture with high-performance and low-memory in JPEG2000. Performance of integer-to-integer DWT is improved by extending precision of raw image data and preserving efficiently fractional bits of transformed coefficients in lifting steps. The proposed scheme outperforms existing methods in achieving higher image compression quality. Furthermore, in order to get over huge wavelet coefficient memory, a memory design based on code block strips is proposed. Efficient reusing and scheduling of memories with code block strips in subbands reduce hardware resources both in storage and power. When a 512×512 image is decomposed and the size of code block is chosen as 32×32, the memory reduction is 80%, compared with directly buffering all the coefficients in the transformed image. The architecture has passed FPGA verification and the synthesized clock frequency can be up to 150MHz.