The project will build on a solid foundation of existing work by Baldwin's AI group in Bristol and the LlFE project (Japan), forming a high-level co-operative system to create declarative models for image understanding. A model will be generated from a set of images, and will identify objects using hierarchical methods. At the bottom of the hierarchy, image processing provides features such as edges. These are combined using fuzzy techniques into meaningful line segments, then into higher level features such as simple and compound shapes. The model may be altered or refined by human expertise or by genetic programming.
The final system will be applicable to different areas of intelligent signal processing. As a demonstrator we will use a library of cervical smear images (created during collaboration between the AI Group and Southmead Hospital, Bristol). The system could partially automate screening of cervical smears-enabling more frequent screening at lower cost.