Precision for 2-D discrete wavelet transform processors

The 2-D discrete wavelet transform (DWT) suits image processing applications well, allowing for excellent compression. Many architectures have been proposed to perform the 1, 2 and 3-D DWT, but few address the precision necessary to ensure perfect reconstruction. The goal of this work is to experimentally determine the precision needed to store the results of a 2-D DWT without introducing round-off error via the 2-D DWT processor.