Result-Biased Distributed-Arithmetic-Based Filter Architectures for Approximately Computing the DWT

The discrete wavelet transform is a fundamental block in several schemes for image compression. Its implementation relies on filters that usually require multiplications leading to a relevant hardware complexity. Distributed arithmetic is a general and effective technique to implement multiplierless filters and has been exploited in the past to implement the discrete wavelet transform as well. This work proposes a general method to implement a discrete wavelet transform architecture based on distributed arithmetic to produce approximate results. The novelty of the proposed method relies on the use of result-biasing techniques (inspired by the ones used in fixed-width multiplier architectures), which cause a very small loss of quality of the compressed image (average loss of 0.11 dB and 0.20 dB in terms of PSNR for the 9/7 and 10/18 wavelet filters, respectively). Compared with previously proposed distributed-arithmetic-based architectures for the computation of the discrete wavelet transform, this technique saves from about 20% to 25% of hardware complexity.