Building a modified block matching kernel based on Wave Atom transform for efficient image denoising

Abstract

The block-matching and 3D filtering strategy is a recent adequate solution for different problems related to low-pass image filter; however, this technique presents some drawbacks. In fact, handling noise with variable density and average or Gaussian noise with high densities may limit the performance of this approach. To overcome these limitations and increase the performance of the original BM3D approach, we propose in this work a modified BM3D technique based on wave atom transform with adaptive threshold called BM3D-AT. Experimental results and simulations revealed that the proposed technique is more efficient in removing different types of noise from different images, hence achieving outstanding performances.