Colloquium "Fast Least Absolute Deviation Adaptive Wavelet Filter Bank"

Abstract:
Sparse representation of signals and images is a key for many applications. Sparseness is often achieved through linear transforms with good energy compaction property. But, the shortest description demands signal dependent bases, which are usually related to high computational complexity and non-local character of decomposition functions. Hence, instead of optimal bases, efficient linear transforms with compactly supported base functions are often used. In this presentation, a widely used wavelet transform is connected with local adaptation to achieve maximum sparseness. The adaptive wavelet filter bank retains good properties of fixed wavelets while adapting to the local signal features. Adaptation is robust, conducted at each sample according to the least absolute deviation (LAD) criterion. Hence, it provides for more sparse representation of the large class of input signals. It results in more accurate adaptation on sudden changes of signal statistics, and it has low computational complexity. The proposed scheme will be used for signal and image denoising and image compression. 

CV:
Ana Sović is a Research Assistant - senior assistant at the University of Zagreb, Faculty of Electrical Engineering and Computing, Croatia. She received her Diploma degree and PhD degree in electrical engineering from the University of Zagreb in 2006 and 2012 respectively. Her research interest includes e-learning and advanced signal and image processing. She is a member of IEEE and European Association for Signal Processing (EURASIP).