Title: Image denoising, from inverse problems to deep learning
Abstract:
In the signal and image processing field, denoising has always been a hot research topic for many applications and over several time periods where methodological tools significantly evolved. From filtering and transform-based techniques to inverse problems and Bayesian models, the question of handling noise is still animating the research field in signal and image processing. Recently, the emergence of deep learning as new paradigm where learning from data is proposed as alternative to mode-based techniques, is transforming the denoising literature. In this talk, we will go for a trip to discover how denoising is handled over decades up to the era of deep learning.
Biography:
Lotfi Chaari is full professor with Toulouse INP where he worked as associate professor from 2012 to 2024. He is a researcher with the IRIT lab (MINDS team, SI department). His research is focused on model-based and data-driven approaches for signal and image processing/analysis. With a large background in inverse problems and Bayesian methods, he mainly works on developing hybrid optimization algorithms, combining variational and Bayesian tools, for sparse deep neural networks learning where all parameters are automatically estimated from the data. These methods are applied to biomedical signal and image processing, as well as to remote sensing. He also works on light machine learning techniques for data analysis in those fields. The handled data is generally multi-dimensional with temporal/spectral dimensions.