Denoising

During acquisition and transmission, images are often corrupted by additive noise. The aim of a denoising algorithm is to reduce the noise level, while preserving the image features. There is a big diversity of estimators used as denoising systems. One may classify these systems into two categories: those directly applied to the signal and those who use a WT before processing. In fact, Donoho and Johnstone introduced the word denoising in association with the wavelet theory (http://www-stat.stanford.edu/~donoho). The multiresolution analysis performed by the WT has been shown to be a powerful tool to achieve good denoising. In the wavelet domain, the noise is uniformly spread throughout the coefficients, while most of the useful information is concentrated in the few largest ones (sparsity of the wavelet representation). The corresponding denoising methods consist of three steps: 1) the computation of the forward WT; 2) the filtering of the wavelet coefficients; and 3) the computation of the IWT of the result obtained. Consequently, there are two tools to be chosen: the WT and the filter. In what concerns the first choice, we currently use a new implementation of the Hyperanalytic Wavelet Transform (HWT). Concerning the second choice, numerous nonlinear filter types can be used in the WT domain. A possible classification is based on the nature of the noise-free component of the signal to be processed. Basically, there are two categories of filters: those built assuming only the knowledge of noise statistics (a nonparametric approach) and those based on the knowledge of both signal and noise statistics (a parametric approach). For more information, see the The use of wavelet theory for decision making project page, Director: Alexandru Isar, Members: Ioan Nafornita, Sorin Moga, Corina Nafornita, Maria Kovaci, Marius Oltean, Ioana Firoiu.



The denoising is a pretreatment required for EKG processing (segmentation in this example).


There are images (SAR, SONAR, biomedical) perturbed by speckle noise (left). The result of a denoising procedure is presented in the right image.

Selected Publications:


Intelligent Signal Processing Centre
Copyright © 2010