![]() Traditionnellement, cette connaissance a priori de l'objet observé a souvent été réservée au traitement du signal en fin de la chaine d'acquisition. Nous verrons que les barrières deviennent de plus en plus poreuses entre ces domaines et que pour la conception de nouveaux capteurs, l'imagerie doit pouvoir utiliser la structure de l'objet étudié. Nous présenterons un panorama des différentes techniques qui permettent de capter l'information partant du capteur conventionnel en imagerie directe jusqu'aux techniques de Machine Learning utilisées dans l'apprentissage de données de très grandes dimensions. This expository talk will use, in part, material from and examples featured in. ![]() We will see in particular how the appearance of sharp phase transitions linked to how much information can be transmitted by the sensor, now provides new rules on the conception and calibration of conventional and non-conventional imaging sensors. In fact, the appearance of these tools has provided the applied and sometimes the not-so-applied mathematicians or the signal processing experts a direct say in the conception of new detectors. For the past ten years, the use of a priori knowledge, such as sparsity, low-rankedness or more generally the manifold in which the signal “lives”, has enabled the development of new mathematical approaches and attendant numerical solution techniques. Traditionally, this task used to be the domain of signal processing experts at the end of the data acquisition chain. In particular, we will show that the traditional barriers between these fields are becoming porous and that in order to conceive new sensors, the use of the structure of the objects being observed as an a priori, is becoming central. We will present a panorama of sensing techniques from direct imaging all the way to Machine Learning.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |