A new image denoising algorithm with multiscale products

Mihcak et al. proposed a locally adaptive window-based denoising method using maxsimum likelihood (LAWML) with low complexity. However, LAWML was based on decimated wavelet transform (DWT) without taking account of the interscale dependencies. In this paper, we propose a variant of LAWML, namely MPLAWML, based on multiscale products. We improve LAWML by extending DWT to undecimated wavelet transform (UWT) and multiplying the adjacent wavelet subbands to exploit the wavelet interscale dependencies. In the multiscale products, edges are enhanced and noise is weakened. Thereafter, a product threshold is calculated for each product subband and is used on the products coefficients to identify significant features. Then LAWML is applied to process those wavelet coefficients, which are greater than the corresponding products thresholds. Experiments show that the proposed algorithm has more robustness to noise, achieves better visual effects than LAWML and has competitive performance compared with the state-of-the-art wavelet-based denoising algorithms.