Wavelet domain image and video denoising and applications in medical imaging
Aleksandra Pizurica


This talk focuses on Bayesian wavelet domain image denoising and its applications in video processing and medical imaging.
The first part of the talk presents the core of our denoising approach where we estimate the probability of the presence of image features of interest in the wavelet domain. Modeling a signal (or a feature) of interest as a noise-free coefficient component that exceeds a certain magnitude, we propose a specific prior on the noise-free subband data that is a mixture of generalized Laplacians. We develop the least squares estimator under the assumed prior and we discuss its properties and relationships to related estimators. A local spatial adaptation of our estimator is achieved by conditioning the probability of signal presence on a so-called local spatial activity indicator. In case of multi-band images (like color and multi-spectral images) we define a joint spatial and spectral activity indicator, which takes into account the spatial correlation of the wavelet coefficients as well as their "spectral" correlation across different image bands.
In the second part of this talk we review some applications. First we present recent results in video denoising, where we combine spatial denoising with motion compensated recursive temporal denoising in the wavelet domain. We also discuss real time hardware implementations of the developed algorithms. The last part of the talk addresses medical imaging applications. We present our results in denoising and analysis of ultrasound and magnetic resonance images.

Aleksandra Pizurica was born in Novi Sad, Yugoslavia, on September 18, 1969. In 1994, she received the Diploma degree in electrical engineering from the University of Novi Sad, Yugoslavia, in 1997 the M.Sc. degree in telecommunications from the University of Belgrade, Yugoslavia and in 2002 the Ph.D. degree from the Ghent University, Belgium. Currently she is a postdoctoral researcher at the Department of Telecommunications and Information Processing of the Ghent University. Her research interests include statistical image modelling, Bayesian estimation methods, multimedia applications, medical imaging and remote sensing.

back to the list of talks