A branch and bound decision tree Bayes classifier for robust multi-font printed Chinese character recognition

A branch and bound classifier is proposed as an m-ary decision tree with each node representing a set of disjoint classes. Associated with each set is a space S and an estimate of the maximum likelihood of any x in S belonging to a class of the set. By comparing this estimate with the best-likelihood-found-so-far for x, it can be decided if the node is worth visiting. This classifier is applied to recognize 4879 classes of multi-font printed Chinese characters with practically the same recognition rate (98%), but in 5% of the time required, when compared with a full-search Bayes classifier