Search

This book explains how to measure the security of a watermarking scheme, how to design secure schemes but also how to attack popular watermarking schemes. This book gathers the most recent achievements in the field of watermarking security by considering both parts of this cat and mouse game. This book is useful to industrial practitioners who would like to increase the security of their watermarking applications and for academics to quickly master this fascinating domain.

Ce chapitre présente le problème de la distribution de documents à des utilisateurs qui ne sont pas de toute confiance. Il propose un modèle mathématique de la personnalisation des documents, des attaques des traîtres (utilisateurs malhonnêtes fuitant le document) et de la manière de les identifier à partir du document pirate. Ce modèle permet de mettre en valeur les conditions nécessaires et suffisantes pour accuser avec raison les traîtres. Mots-clés : traçage de traîtres, fingerprinting, code de Tardos, théorie des jeux. DOI : 10.51926/ISTE.9026.ch6

L’apprentissage automatique utilisant des réseaux neuronaux profonds appliqués à la reconnaissance d’images fonctionne extrêmement bien. Néanmoins, il est possible de modifier intentionnellement et très légèrement les images, modifications quasi invisibles à nos yeux, pour leurrer le système de classification et lui faire classer dans une catégorie visuelle erronée tel ou tel contenu. Ce chapitre dresse un panorama de ces attaques intentionnelles, mais aussi des mécanismes de défense visant à les déjouer. Mots-clés : images adverses, classification, attaques, défenses, leurrage. DOI : 10.51926/ISTE.9026.ch2

We consider in this thesis the use of digital watermarking in the copy protection framework for consumer electronics devices. We describe first the copy protection issue. Then, we build the global system with elementary securityparts such as encryption and digital signature. Yet, contrary to common approaches, the role of the watermark hasbeen reduced to the minimum: it is only a flag warning the devices that the content is protected. It is a kind ofsecond line of defence. Watermarking for copy protection is difficult due to two facts: the protected contents arewatermarked with the same key and the pirates have access to a watermark detector. Three kinds of attacks stemfrom these facts: the watermarked contents only attack, the original/watermarked contents attack and the chosenwatermarked contents attack. Even if we manage to reduced the capacity to one bit, the choice of a watermarkingtechnique is still difficult: an analysis shows that classical spread spectrum techniques do not provide a sufficientlyhigh level of security for this application. This is the reason why we invent a new class of methods known asasymmetric watermarking. This provides high security level but requires a bigger amount of data to detect thewatermark. To boost the detector, we take advantage of the side information at the embedding stage to optimisethe watermark detection. This gives birth to another new method so-called JANIS. For a small loss in the securitylevel, the detector is much more efficient. To conclude, these new methods may justify the Kerckhoffs principle in watermarking for copy protection.