 |
| |
 |
Objectives |
|
|
|
|
CASAM introduces the concept of computer-aided semantic annotation to accelerate the adoption of semiautomatic multimedia annotation in the industry. The project is unique in that it facilitates the synergy of human and machine intelligence to significantly speed up the task of human-produced semantic annotation of multimedia content.
The proposed approach is motivated by the limitations of current efforts to fully automate multimedia content annotation, i.e. the immature machine-only annotation and the cost-prohibitive manual annotation.
The balance of the human and machine effort will depend on the context and complexity of the multimedia document. Therefore, it is particularly important that this balance is not predetermined and fixed, but rather adaptive to the situation. 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.
The CASAM project will contribute to the state of the art by:
- Providing the technology for knowledge-driven multimedia analysis. The novel characteristics of the approach include the reassessment of context and concepts based on user feedback and new knowledge. The multimedia analysis procedure will be able to refine its finding after positive or negative feedback resulting in a more robust approach and better handling of uncertainty.
- Developing novel methods of reasoning for multimedia interpretation that are built around information exchange between the multimedia analysis procedure and the human. The reasoning subsystem will be able to identify, explicitly specify and communicate the information that it needs in order to reach the targeted annotation detail and level.
- Designing unique intelligent human-computer interaction methods that act in order to maximize the expected information gain from the user’s input while at the same time provide a cooperative environment to the human and underlying machine reasoning to guide the knowledge aggregation process.
Apart from the general project objectives there are also second level, measureable objectives:
- Accurate extraction of context and key objects from multimedia content via video / image analysis.
- High speech-to-text efficiency in identifying keywords even in noisy environments.
- Language-independent text processing and population of ontology.
- Efficient human-machine interface and minimization of human effort.
- Increased overall annotation speed.
|
|
|
 |
|
|
 |
| Events |
 |
|
 |
| |
|
Follow our discussions and news at:
|
|
 |
|