**3.8 Novelty detection**

The objective is to develop a CBR-based novelty detection method based on a combination of statistical and similarity-based methods able to handle the different multimedia data types.

The method will perform novelty detection and handling of the novel cases for immediately reasoning and new model built up and should consider the incremental nature of the data. It will update the models incrementally based on MML and MDL methods.

• State of art on statistical novelty detection methods applicable to multimedia data types.

Different statistical model methods will be studied and evaluated for the different multimedia data types. The methods that are applicable to the data types will be selected.

• Model development for novelty detection classification for multimedia data types

Taking into account the outcome of Task 8.1, new methods for the different multimedia data types will be developed and implemented.

• Incremental model learning mechanism based on MDL or MML

The incremental learning mechanism for the updating of the models and feature description should be developed based on statistical learning methods such as MML or MDL.

• Incremental data collection mechanism and data base structure

Development of the data base structure for the different multimedia data types.

The incremental data collection mechanism will be developed, taking into account the data base structure.

• Task management mechanism between the model-based and the similaritybased unit

The task management mechanism that can handle the interaction and task division between the model-based module and the similarity-based module will be developed, implemented and evaluated.

• Integration into the CBR unit

The developed novelty detection unit will be integrated into the CBR unit and tested.
