ABSTRACT
This chapter describes a rather general geometric model of object recognition, and discusses the implications of this model for the associated problem of object representation. It provides particular coding strategies which have proved successful in a machine recognition context. The chapter examines the associated problems of object recognition and representation in terms of the task demands, and reviews a number of existing theories from this viewpoint. It shows that how the task demands suggest a differentiable manifold model of recognition, and explores the general implications of this assumption. The chapter also explores the implications of manifold model for more general recognition tasks, suggesting that the model is not specific to faces, and indeed moves smoothly between intra-class and inter-class discrimination, yielding what seems to be a parametrized version of geon-based recognition. It also discusses how a manifold model of face recognition leads to a re-examination and extension of existing models of face recognition.
