EBCOT-based discrete wavelet transform scheme with row-overlapped Z-scan

For JPEG 2000 implementation in a low-cost circuit, it is important to reduce the sizes of the internal buffers and the numbers of external memory accesses, especially for DWT related parts. We present an EBCOT-based DWT technique that can produce the DWT coefficients strictly in the same sequence as to code them in EBCOT (embedded block coding with optimized truncation). The large buffer for coefficient transfer can be removed completely. With a specific row-overlapped Z-scan pattern to carry out DWT, the sizes of the internal buffers are reduced to only 1.7% or 2.9% at the cost of memory access increase of 2.7% or 5.5% for (5,3) or (9,7) wavelet filters respectively.