Journal of South China University of Technology (Natural Science Edition) ›› 2009, Vol. 37 ›› Issue (5): 38-42.

• Electronics, Communication & Automation Technology • Previous Articles     Next Articles

High-Performance and Low-Memory Architecture of Wavelet Transform for JPEG2000

Guo Jie  Wu Cheng-ke  Li Yun-song  Ma Jing   

  1. State Key Laboratory of Integrated Service Networks, Xidian University, Xi'an 710071, Shaanxi, China
  • Received:2008-05-15 Revised:2008-11-12 Online:2009-05-25 Published:2009-05-25
  • Contact: 郭杰(1982-),男,博士生,主要从事图像压缩编码及其硬件实现研究. E-mail:jguo@mail.xidian.edu.cn
  • About author:郭杰(1982-),男,博士生,主要从事图像压缩编码及其硬件实现研究.
  • Supported by:

    国家自然科学基金重点项目(60532060);国家自然科学基金资助项目(60507012,60802076);西安电子科技大学博士创新基金资助项目(创05025)

Abstract:

In order to improve the performance of integer discrete wavelet transform (DWT) and the compression quality of images, a DWT architecture with high performance and low memory is proposed for JPEG2000. This architecture extends the precision of raw image data and effectively preserves the fractional bits of transformed coeffi- cients in lifting steps. It adopts a memory scheme of wavelet coefficients based on code block strips to reuse and schedule the memories with code block strips in sub-bands, thus reducing the hardware resources in terms of sto- rage and power consumption. Experimental results show that the proposed scheme effectively improves the compres- sion quality of images, and that, by decomposing a 512 × 512 image using a code block in the size of 32 × 32, the memory of the proposed architecture is only 20% that of the scheme directly buffering all the coefficients in the transformed image. The proposed architecture has passed the FPGA verification and the synthesized clock frequency of it is up to 150 MHz

Key words: JPEG2000, discrete wavelet transform, lifting scheme, code block, VLSI circuits