Contribution to the construction of fingerprinting and watermarking schemes to protect mobile agents and multimedia content

Resumen

The main characteristic of fingerprinting codes is the need of high error-correction capacity due to the fact that they are designed to avoid collusion attacks which will damage many symbols from the codewords. Moreover, the use of fingerprinting schemes depends on the watermarking system that is used to embed the codeword into the content and how it honors the marking assumption. In this sense, even though fingerprinting codes were mainly used to protect multimedia content, using them on software protection systems seems an option to be considered. This thesis, studies how to use codes which have iterative-decoding algorithms, mainly turbo-codes, to solve the fingerprinting problem. Initially, it studies the effectiveness of current approaches based on concatenating tradicioanal fingerprinting schemes with convolutional codes and turbo-codes. It is shown that these kind of constructions ends up generating a high number of false positives. Even though this thesis contains some proposals to improve these schemes, the direct use of turbo-codes without using any concatenation with a fingerprinting code as inner code has also been considered. It is shown that the performance of turbo-codes using the appropiate constituent codes is a valid alternative for environments with hundreds of users and 2 or 3 traitors. As constituent codes, we have chosen low-rate convolutional codes with maximum free distance. As for how to use fingerprinting codes with watermarking schemes, we have studied the option of using watermarking systems based on informed coding and informed embedding. It has been discovered that, due to different encodings available for the same symbol, its applicability to embed fingerprints is very limited. On this sense, some modifications to these systems have been proposed in order to properly adapt them to fingerprinting applications. Moreover the behavior and impact over a video produced as a collusion of 2 users by the YouTube’s s ervice has been s tudied. We have also studied the optimal parameters for viable tracking of users who have used YouTube and conspired to redistribute copies generated by a collusion attack. Finally, we have studied how to implement fingerprinting schemes and software watermarking to fix the problem of malicious hosts on mobile agents platforms. In this regard, four different alternatives have been proposed to protect the agent depending on whether you want only detect the attack or avoid it in real time. Two of these proposals are focused on the protection of intrusion detection systems based on mobile agents. Moreover, each of these solutions has several implications in terms of infrastructure and complexity.