A novel approach is proposed, which allows for an efficient reduction of the amount of visual data required for representing structural information in the image. This algorithm is tolerant to minor structural changes and can be used for automatic face recognition. The approach is based on a multistage architecture, which investigates partial clustering of structural image components. The initial grey-scale representation of the input image is transformed into a structural representation, so that each image component contains information about the spatial structure of its neighbourhood. The output result is represented as a pattern vector, whose components are computed one at a time to allow the quickest possible response. The input pattern is identified as the best match between the output pattern vector and the model vectors from the database.
展开▼