A principal focus of this theme is the design, implementation and analysis of high performance classifier systems, and especially structures based on multi-classifier configurations. We have a long track record of excellence in developing structures which integrate information from a variety of different classifiers, thus building on the strengths of individual approaches and avoiding their weaknesses. Our work has recently focused on the optimisation of multi-classifier systems, and we have developed a powerful approach based on the use of genetic algorithms to optimise structures while, in parallel, we are developing a theoretical framework which can inform system design in a more generic way. We are also interested in issues surrounding dynamic optimisation. All this work is of a generic nature, but also has immediate applicability in a variety of practical scenarios, and thus underpins aspects of several of the other research themes in the Group.
Related work in this theme includes some fundamental work on feature extraction and feature selection/optimisation for classification processing. The focus of this work has been on the use of colour and texture-based features to support the implementation of high performance vision systems, and there are clear links to the exploitation of such features in multi-classifier configurations. Some of this work is moving into the area of 3-D processing for vision, with applications in face recognition a particular interest.
This research theme has attracted significant external funding, with our generic research on multi-classifier design and optimisation supported for more than ten years through a continuing series of EPSRC research grants. Both the generic and application-oriented aspects of this work have stimulated important publications in the scientific literature, and have been reported at major international Conferences and Workshops. One important outcome has been the publication of a detailed literature review as a resource for the whole research community, and the Group has organised Special Sessions on this theme in several major conferences.