Tiled Interleaving for Multi-Level 2-D Discrete Wavelet Transform

This paper presents a new architecture of 2D discrete wavelet transform (DWT) proposed for JPEG 2000. In the proposed architecture, the image is segmented into tiles each of which is sequentially processed to minimize the size of buffers required to process 2D DWT, and multi-level DWTs are interleaved to reduce the size of the repeat buffer drastically. Compared to the conventional architecture, the overall memory size is reduced by 85% and 92% for 256 times 256 and 512 times 512 images, respectively. The proposed DWT processor needs only 5kB memory for 256 times 256 images, and operates at 250MHz in 0.25-mu technology.