Robust Expert Ranking in Community-based Fake Multimedia Detection Systems

Recently, many images found on the Internet and social network sites like Flickr are fake multimedia, primarily made with various multimedia editing  techniques. Detection and management of the fake multimedia in large web-based social networks or P2P networks is a challenge. Traditional faked multimedia identification techniques such as digital watermarking and digital forensics lack the robustness and scalability; content-based approaches using low-levelfeatures (color, shape and texture) still do not bridge the semantic gap to identify fakes. Exploiting Web 2.0 and community approaches together with content based multimedia retrieval is the proposed solution.

Identifying the fake multimedia with the help of users required knowledge whether the users in the community are trusted and what expertise they have in a given field.

The task of this thesis is the development of an expert ranking algorithm for a community-based fake multimedia detection system based on trustworthiness and expertise of community agents and to offer countermeasures against compromising this algorithm, assuring its robustness. Also, a multimedia similarity search service will be implemented. The service will use the MQP of the files to detect similar or identical media files. The implemented system will be at the end evaluated using a system for collaborative fake media detection in a real-time distribution network developed at the i5 chair and the robustness of the expert ranking algorithm will be checked using robustness evaluation metrics.