Jimmy TEKLI : "Design and evaluation of image anonymisation techniques and relational data streams using machine learning approaches a BMW Group"

Abstract : Individual’s privacy and anonymity is becoming highly critical in our data-driven world due to the vast amount of data being generated and processed daily (e.g., Industry 4.0). Data anonymization is the process of creating anonymous information, namely information which does not relate to an identified or identifiable natural person in such a manner that the data subject is not or no longer identifiable. Privacy regulations compel data-driven companies to guarantee a level of anonymization that requires “irreversibility preventing identification of the data subject”, taking into account all the means “reasonably likely to be used” for identification. Therefore, we (i) propose and implement several anonymization techniques and tools in the context of images and relational data streams and (ii) assess the robustness of these techniques by simulating adversaries with different knowledge and several attacking capabilities. In the first contribution, we design and implement an anonymization tool that localizes identifying/sensitive features in images/videos via Deep Learning DL-based localization techniques (i.e., semantic segmentation) and obfuscates it accordingly via pixelating, blurring, or masking. In the second contribution, we propose a recommendation framework that evaluates the robustness of image obfuscation techniques and recommends the most resilient obfuscation against adversaries executing DL-assisted attacks (e.g., restoration or recognition-based attacks). In addition, three threat levels are studied thoroughly based on the adversary’s knowledge (e.g., background knowledge). In the third contribution, we empirically demonstrate how adversaries can remedy their lack of knowledge and leverage their attacking capabilities, against obfuscated facial images, by collaborating via Federated Learning. [...]

Jury Composition  :

M. Raphaël COUTURIER Full professor Université Bourgogne Franche-Comté - PhD Director
M. Richard CHBEIR Full professor Universite de Pau - Reporter
M. David ELIZONDO Full professor The De Montfort University - Reporter
M. Bechara AL BOUNA Full professor Université Antonine - PhD Co-Director
M. Lionel BRUNIE Full professor Laboratoire d'Informatique en Images et Systèmes d'Information (LIRIS) -  Reviewer
M. Chris CLIFTON Full professor Purdue University - Reviewer
M. Mohamed NASSAR Assistant professor University of New Haven - Reviewer

Localization : Room F144 - IUT Belfort-Montbéliard 19, Avenue du Maréchal Juin 90000 BELFORT