Seminars


 

 

4. [29/12/2021] Talk by Steve McLaughlin from School of Eng and Physical Sciences, Heriot-Watt University

Title: Expectation-Propagation methods for imaging inverse problems

Summary: In this talk I will introduce EP based methods and their application to a range of inverse problems in imaging. The talk will introduce the basic ideas involved in EP based methods and then illustrate the application to a linear unmixing problem in hyper spectral imaging. Then the problem of image restoration using patch-based priors in an EPO framework will be presented. Finally some discussion in estimating the hyper parameter will be presented.

3. [04/2021] Talk by Duong Hung PHAM from IRIT lab

Title: Blind deconvolution-based clutter suppressionfor vascularization imaging

Summary: In this talk, I present the problem of high-resolution Doppler blood flow estimation from an ultrafast sequence of ultrasound images. Formulating the separation of clutter and blood components as an inverse problem has been shown in the literature to be a good alternative to spatio-temporal singular value decomposition (SVD)-based clutter filtering. In particular, a deconvolution step has recently been embedded in such a problem to mitigate the influence of the experimentally measured point spread function (PSF) of the imaging system. Deconvolution was shown in this context to improve the accuracy of the blood flow reconstruction. However, measuring the PSF requires non-trivial experimental setups. To overcome this limitation, we propose herein a blind deconvolution method able to estimate both the blood component and the PSF from Doppler data. Numerical experiments conducted on simulated and in vivo data demonstrate qualitatively and quantitatively the effectiveness of the proposed approach in comparison with the previous method based on experimentally measured PSF and two other state-of-the-art approaches.

2. [03/2021] Talk by Julien Lesouple (thanks Julien!) from TeSA lab

Title: Incorporating expert feedback into anomaly detection using support vector machines

Summary: Anomaly detection consists of detecting elements of a database that are different from the majority of normal data. The majority of anomaly detection algorithms is adapted to unlabeled datasets. However, in some applications, labels associated with a subset of the database (coming for instance from expert feedback) are available, providing useful information to design the anomaly detector. This presentation introduces a semi-supervised anomaly detector based on support vector machines, which takes the best of existing supervised and unsupervised Support Vector Machines (SVM) algorithms.The first part is dedicated to the introduction of supervised SVM and their application to unsupervised anomaly detection, then we will explain how anomaly dectection can be performed using SVM with partially labelled datasets and how this can be applied to anomaly detection with expert feedback, and we will finish by presenting current studies and potential future works

1. [01/2020] A talk of Mario Figueiredo on Wednesday 8 at 4 PM, at TeSA

Title: Three Recent Short Stories About Image Denoising

Mário Figueiredo,  Instituto de Telecomunicações, Instituto Superior Técnico University of Lisbon Portugal

In this talk, I will review three recent proposals in the area of patch-based image denoising. The first one is a simple post-processing technique for Poisson denoisers, based on an elementary application of classical linear minimum mean squared error (LMMSE) estimation, and which is able to squeeze a few extra tenths of dB of ISNR from several state-of-the-art Poisson denoisers and to produce better-looking images. The second part of the presentation describes how external non-local means (NLM) denoising can be seen as an important sampling approach to computing MMSE patch estimates, which opens the door to using NLM with arbitrary noise models. The third and final part of the talk is about denoising interferometric (phase) images using multi-resolution windowed Fourier filtering, guided by Stein’s unbiased risk estimate (SURE), which outperforms previous state-of-the-art methods for this problem.