ISP Seminars

Previous Seminars

Diffusion MRI and DKI to evaluate the axonal degeneration in vivo: general framework and focus on some processing tools
Stéphanie Guérit (ICTEAM/ELEN, UCL)
22 January 2014
A lot of incidents in everyday life can lead to spinal cord injuries: motor vehicle accidents, falls, sport injuries, etc. In some cases, the consequences are superficial or reversible but they can also have an important impact on the quality of life of the patients by the permanent loss of functionality. At this time, the process of axonal degeneration in the central nervous system (CNS), responsible for this functional loss, is not well known. The current clinical methods to evaluate the extent of the damage are based on the sensorial perception of the patient and its ability to contract some of his muscles. There exists no objective method based on the observation of physiological processes.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 22 January 2014 at 10:00 (40 min.)

Collaborative algorithms, bees & NodeJS
Jérôme Plumat (ICTEAM/ELEN, UCL)
20 November 2013
Collaborative algorithms are very powerful tools to rapidly investigate a huge unknown space of solutions. The presentation introduces a very simple collaborative problem: path finding in graph. The proposed solution is based on bees and analogy to the pollen gathering. An algorithm is presented in order to take into consideration information collected by others and incorporates it in a prior model. A very brief introduction to efficient programming tools aims you to realize how such tools may be easily implemented.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 20 November 2013 at 10:00 (30 min.)

Compressed Sensing of Low Complexity High Dimensional Data: Application to Hyperspectral Imaging
Kévin Degraux (ICTEAM/ELEN, UCL)
6 November 2013
In numerous applications, acquiring and processing High Dimensional (HD) data is challenging. 3-dimensional imaging and in particular Hyperspectral Imaging (HSI) is an example where acquiring a whole volume made up of bilions of voxels is a hard task. Fortunately, in that case and in many others, the data volume is most often highly structured and full of redundancy. This framework is the ideal field of application for Compressed Sensing (CS). That theory allows to exploit these redundancy priors to blindly compress data at the analog acquisition stage. Provided that the right priors are used and that the CS sensor model respects some mathematical properties, we can strongly reduce the sampling rate (w.r.t. Nyquist rate) or equivalently increase target resolution and focus on reconstruction afterwards. In this talk, I will describe at a high level of abstraction every step of the CS acquisition and reconstruction of HD data and in particular Hyperspectral images with classical and more exotic priors like low rank or hybrid models.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 6 November 2013 at 10:00 (45 min.)

Continuous parameter estimation from compressive samples
Prasad Sudhakar (ICTEAM/ELEN, UCL)
9 October 2013
For several applications, it is sufficient only to extract a few parameters of a signal, from its compressive measurements, instead of having a full reconstruction, thereby saving a lot of computational effort. Often, the underlying parameters that characterize the signal are drawn from a continuous space. However, the standard compressive sensing formalism is discrete in nature and hence the parameter estimates are confined to a predefined grid. In order to go off the grid, one has to exploit the underlying continuous model and perform either gradient descent or interpolation. In this talk, I will consider a very simple signal model and describe how to estimate continuous parameters from compressive samples.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 9 October 2013 at 10:00 (45 min.)

Optimization of MRI reconstruction by constraining the minimization of TV norm
Damien Jacobs (ICTEAM/ELEN, UCL)
25 September 2013
In high field MRI, the SNR is significantly increased and makes possible to improve the voxel resolution. A high number of data's can be acquired on the patient and should help his diagnosis by the physician. The timing restriction to scan the patient is an obligation in the medical institutions. Getting a high resolution with a limited time scan is a real challenge in MRI reconstruction. Also, high rf stimulation and gradient amplitudes are restricted in high field MRI to avoid nerves stimulation. These constraints make harder the sampling and the acquisition of data's. The sampling of data's in MRI coupled to the MRI reconstruction is a key point and its interest is growing with the presence of high field MRI. In this ISPS, a well-known approach of convex optimization by minimizing Total-Variation norm will be presented in the case of MRI reconstruction. Simulation of data's and real phantoms acquired on the 11.7 T MRI scan will be presented.

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 25 September 2013 at 10:00 (45 min.)

Computational inverse scattering and TV regularization: A cocktail party explanation
Augustin Cosse (ICTEAM/ELEN, UCL)
4 September 2013
Imaging by inverse wave scattering is a multidisciplinary area implying concepts from both physics, mathematics and signal processing. Current approaches to the problem fall into two categories : inverse obstacle problem and inverse medium problem. In the former, the scattering object is a homogeneous obstacle one wants to determine from the field on the boundary. In the latter, the scattering object is an inhomogeneous medium with respect to some physical parameters. The inverse problem then consists in estimating those parameters from the field on some boundary. The talk will be focusing on the second problem. Such problem is usually not only non-linear but can also be severely ill-posed. Moreover, obtaining accurate solutions usually requires huge datasets. In the light of recent developments in the compressive sensing (CS) theory together with the increased computing capabilities, we try to obtain high accuracy images of an object with respect to variations in the physical parameters. This is done using somehow advanced understanding of the wave propagation physics while at the same time looking for a decrease in the complexity of current methods.

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 4 September 2013 at 14:00 (45 min.)

Detection of facial expressions and computer graphics animation
Kaori Hagihara (ICTEAM/ELEN, UCL)
19 June 2013
I will introduce my current project collaborating with a 3D Computer Animator. In 3D Computer Animation, the production stage involves building, rigging and texturing models, animating characters, and setting up and lighting scenes. One of the task, animating characters, requires huge amount of manual manipulations. In order to facilitate this task, limited to facial animations, we are developing a semi-automate system by capturing an actor with a kinect camera. The facial action of the actor will drive facial animation of the character. Technically, the system consists of two parts: detection of facial actions and animation/morphing of a character. The talk will be mainly about the detection of facial expressions using sequences of RGB and depth images from kinect camera.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 19 June 2013 at 14:00 (45 min.)

Training with corrupted labels to reinforce a probably correct team-sport player detector
Pascaline Parisot (ICTEAM/ELEN, UCL)
5 June 2013
While the analysis of foreground silhouettes has become a key component of modern approach to multi-view people detection, it remains subject to errors when dealing with a single viewpoint. Besides, several works have demonstrated the benefit of exploiting classifiers to detect objects or people in images, based on local texture statistics. We train a classifier to differentiate false and true positives among the detections computed based on a foreground mask analysis. This is done in a sport analysis context where people deformations are important, which makes it important to adapt the classifier to the case at hand, so as to take the teamsport color and the background appearance into account. To circumvent the manual annotation burden incurred by the repetition of the training for each event, we propose to train the classifier based on the foreground detector decisions. Hence, since the detector is not perfect, we face a training set whose labels might be corrupted. We investigate a set of classifier design strategies, and demonstrate the effectiveness of the approach to reliably detect sport players with a single view.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 5 June 2013 at 14:00 (45 min.)

Discriminative Label Propagation for Multi-Object Tracking with Sporadic Appearance Features
Amit Kumar K.C. (ICTEAM/ELEN, UCL)
15 May 2013
Given a set of plausible detections, detected at each time instant independently, we investigate how to associate them across time. This is done by propagating labels on a set of graphs that capture how the spatio-temporal and the appearance cues promote the assignment of identical or distinct labels to a pair of nodes. The graph construction is driven by the locally linear embedding (LLE) of either the spatio-temporal or the appearance features associated to the detections. Interestingly, the neighbourhood of a node in each appearance graph is defined to include all nodes for which the appearance feature is available (except the ones that coexist at the same time). This allows to connect the nodes that share the same appearance even if they are temporally distant, which gives our framework the uncommon ability to exploit the appearance features that are available only sporadically along the sequence of detections.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 15 May 2013 at 14:00 (45 min.)

A resource allocation framework for adaptive selection of point matching strategies for visual tracking
Quentin De Neyer (ICTEAM/ELEN, UCL)
8 May 2013
The presentation introduces an object tracking framework based on the matching of points between pairs of consecutive video frames. The approach is especially relevant to support object tracking in close-view video shots, as for example encountered in the context of the PTZ camera autotracking problem. In contrast to many earlier related works, we consider that the matching metric of a point should be adapted to the signal observed in its spatial neighborhood, and introduce a cost-benefit framework to control this adaptation with respect to the global target displacement estimation objective. Hence, the proposed framework explicitly handles the trade-off between the point-level matching metric complexity, and the contribution brought by this metric to solve the target tracking problem. As a consequence, and in contrast with the common assumption that only specific points of interest should be investigated, our framework does not make any a priori assumption about the points that should be considered or ignored by the tracking process. Instead, it states that any point might help in the target displacement estimation, provided that the matching metric is well adapted. Measuring the contribution reliability of a point as the probability that it leads to a crorrect matching decision, we are able to define a global successful target matching criterion. It is then possible to minimize the probability of incorrect matching over the set of possible (point,metric) combinations and to find the optimal aggregation strategy.
Ressources :

Shannon Seminar Room (a105) Place du Levant 3, Maxwell Building, 1st floor -- Wednesday, 8 May 2013 at 14:00 (45 min.)