This talk will summarize some of the research contributions made to the fields of computer vision and machine learning. The work focuses on image data representation for retrieval applications but other tasks such as compression, classification and detection can benefit from the proposed representations.
I will highlight the challenges related to the design of image retrieval systems, and present contributions to image representation in association with the evolution of the field.
Some security aspects will also be adressed: how an adversarial query, maliciously modified to disrupt content-based image retrieval system, can be created in a white-box scenario, as well as the detection of tampered images. Since images are often accompanied by text, multimodal aspects and links between NLP and computer vision will also be discussed.
These contributions will finally be put into the perspective of the actual deep learning landscape, opening several research avenues.
The defense will take place in French.
- Julien VELCIN, Professeur des universités, Univ. Lumière Lyon 2, Rapporteur
- Vincent ORIA, Professor, New Jersey Institute of Technology,Et at-Unis, Rapporteur
- Valérie GOUET-BRUNET, Directrice de recherche, IGN
- Luce Morin, Professeur des universités, INSA Rennes
- Thomas Corpetti, Directeur de recherche CNRS, LETG