Hierarchical Binary Vector Quantisation Classi ers

We report on a hierarchical nearest-neighbor classiier algorithm which we conceived for the recognition of handwritten characters. Distances to all classes are used both as a decision criterion in the classiication hierarchy and for generating class membership coeecients. These likelihood values can be easily integrated in a multi-agent cognitive environment. We introduce a new completely binary version of the k-means cluster algorithm and explain how a highly eecient implementation can be achieved using binary patterns. Performances for large character databases are presented.