Novel one-dimensional and two-dimensional forward discrete wavelet transform 5/3 filter architectures for efficient hardware implementation

Abstract

We implemented a more efficient circuit for one-dimensional (1-D) forward discrete wavelet transform (DWT) 5/3 filter. Our design utilizes processing and memory resources that are wasted in some other state-of-the-art solutions and is at least 33% simpler in terms of used registers, is 17% simpler in terms of used logic elements, has 7% higher maximum operating frequency and has 2% lower total power dissipation than previously published designs. The advantages of our design are achieved by a novel non-stationary filter topology which reuses the same registers for generating both low-pass and high-pass output coefficients, in different time slots, due to feed-forward and feedback paths. Our design is suitable for image compression systems which use 5/3 filter, e.g., JPEG 2000. We also proposed two-dimensional (2-D) DWT 5/3 architecture which uses implemented 1-D DWT filter design. The proposed 2-D DWT architecture outperforms all previously published architectures in terms of required memory capacity, which is at least 20% lower than memory capacity in any other reported solution.