On potential to improve DCT-based denoising with local threshold
Abstract
This paper deals with image denoising under standard assumption of additive white Gaussian noise presence. We study potential of denoising effectiveness for a typical representative of orthogonal transform based techniques - a filter based on discrete cosine transform (DCT) with fixed size blocks. It is shown that its performance depends upon many different factors where threshold setting is of prime importance. Possibility of setting individual threshold for each block is considered and it is demonstrated that it opens a certain perspective for further improving performance compared to known versions of DCT-based filter. Performance comparison is done for several denoising techniques including the benchmark BM3D filter. It is shown that potentially DCT-based filter with properly set thresholds adapted to block properties can perform the same or sometimes even better than BM3D filter. © 2018 IEEE.