|
This proposal introduces the notion of computer-aided semantic annotation of multimedia content. Starting from the acknowledgment of the weak points of fully automatic annotation, and the observed gap between manual and automated annotation approaches, this proposal sets the new goal of combining human and machine intelligence to maximize the performance and benefits in a semi-manual annotation scheme. Therefore, instead of trying to substitute human intelligence, the machine will complement it. Hence the novelty of CASAM lies in the difficult task of online aggregating human and machine knowledge with the ultimate target of minimizing human involvement in the annotation procedure.
In order to achieve its ambitious target, CASAM will move current research efforts towards new directions. Knowledge representation and reasoning will play a central role, providing the semantics of the process. In this area, we need to go beyond current research trends, into a closer interaction with both the multimedia analysis tools and the user, with the aim of optimising annotation performance and minimising the user's overhead. On the side of multimedia analysis, we need a knowledge-driven approach that will be able to focus on the context provided by both human and system knowledge. Finally, in the interaction of the human with the system, we need to optimise the acquisition of required information, through the knowledge inferred by the machine about a particular situation.
Towards its main target, CASAM sets specific goals. These pertain to the increase in annotation speed and accuracy compared to both manual and automated annotation. Therefore, the usability of the derived methods and tools and the real-world performance will signal the success of the project. Progress will be measured by real end-users, participating actively, as consortium partners, in all stages of the development of the project. Thus, feedback will be continually given to the rest of the consortium.
|