The objective of the project is to develop new nonlinear model-based descriptions of images and image sequences. The large amount of image-data in multimedia applications makes it necessary to code the information in a model-based symbolic form. A high measure of compactness in the data representation can be foreseen if the power of nonlinear models is taken into account. Beside the potential for image coding purposes a high-level description is an absolute prerequisite to support the application of higher level functions like model-based browsing and navigation, keying, image sequence interpolation, tracking and finding salient regions, querying (similarity measures, indexing, fuzzy similarity measures).

A high measure of compactness in the data representation can be foreseen if the potential of nonlinear models is taken into account like: polynomial models, motion models, Markov models, set-based models (rank-order, morphological, region growing like watershed etc.), stochastic and chaotic models, genetic algorithms and fuzzy models, canonical frames with invariants, 3D models, models for interpolation. The methods must be applied to still images as well as to image sequences. Another point of interest is data fusion in multimedia applications, i.e. to combine features in different representations (e.g. audio and video) to increase the database for searching and classification.

Current models were mainly developed for image coding purposes. They are rather simple and far away from being optimal and do not contribute to more complex tasks like those needed in image databases. The research will focus on standard video sequences; however, more advanced tasks for future standards like stereo pairs and 3D images (virtual reality) will also be investigated.