Title :
Memory Efficient Multilevel Discrete Wavelet Transform Schemes for JPEG2000
Author :
Linning Ye ; Zujun Hou
Author_Institution :
Sch. of Econ. Inf. Eng., Southwestern Univ. of Finance & Econ., Chengdu, China
Abstract :
When working with the line-based wavelet transform, JPEG2000 needs to buffer many sub-band lines for future block coding. This coding buffer occupies the majority of the memory usage in a JPEG2000 system. In this paper, we first use the multilevel block-based wavelet transform, and then utilize the multilevel stripe-based wavelet transform to realize JPEG2000 coding. The system schemes of the multilevel block-based wavelet transform and the multilevel stripe-based wavelet transform for JPEG2000 are presented. The proposed schemes can effectively control the wavelet coefficient output pattern, and thus reduce the memory usage of JPEG2000 coding buffer. Compared with the line-based wavelet transform, the proposed schemes reduce more than 50% memory usage of the JPEG2000 system and slightly decrease the memory bandwidth. By holding the blocks/stripes in on-chip memory, which do not depend on the image size, the memory bandwidth of the JPEG2000 system can be further significantly reduced. The advantage of the stripe-based wavelet transform over the block-based wavelet transform lies in that it occupies much less internal memory.
Keywords :
block codes; data compression; discrete wavelet transforms; image coding; JPEG2000 coding; block coding buffer; image compression; image size; memory efficient multilevel discrete wavelet transform scheme; memory usage; multilevel block-based wavelet transform; multilevel stripe-based wavelet transform; Discrete wavelet transforms; Image coding; Memory management; Transform coding; Wavelet coefficients; Code block; Discrete wavelet transform; JPEG2000; code block; discrete wavelet transform (DWT); lifting scheme; line-based method;
Journal_Title :
Circuits and Systems for Video Technology, IEEE Transactions on
DOI :
10.1109/TCSVT.2015.2400776