Optimisation des Systèmes Multimodaux pour l’Identification dans l’Imagerie

Abstract

Among the most popular media that have taken an essential place for the development of biometric recognition systems in general and face recognition systems in particular, we find Image. One of the most common uses of images is identification/verification in biometrics, which has seen growing interest in recent years. The effectiveness of identification techniques in imaging is today very strongly linked to strong constraints imposed on the user. A current line of research therefore turns to the management of situations where data acquisition is less constrained. Finally, the use of a single modality is often limited in terms of performance or difficulties of use, why it seems interesting to evaluate the contribution of multi-modality in this context. The objective of the thesis is to carry out a work to pursue a research directed toward the techniques of optimization based on the one hand on the hybrid descriptors and the patches as well as their techniques of fusions, and on the other hand on the Deep Learning (Transfer Learning). We are particularly interested in the image of faces and our approaches are validated on several universal databases to tackle all the hazards of acquisition and uncontrolled environments.