OPTICAL CHINESE CHARACTER-RECOGNITION USING PROBABILISTIC NEURAL NETWORKS

Building on previous work in Chinese character recognition, we describ e an advanced system of classification using probabilistic neural netw orks. Training of the classifier starts with the use of distortion mod eled characters from four fonts. Statistical measures are taken on a s et of features computed fi om the distorted character. Based on these measures, the space of feature vectors is transformed to the optimal d iscriminant space for a nearest neighbor classifier In the discriminan t space, a probabilistic neural network classifier is trained. For cla ssification we present some modifications to the standard approach imp lied by the probabilistic neural network structure which yields signif icant speed improvements. We then compare this approach to using discr iminant analysis and Geva and Sitte's Decision Surface Mapping classif iers. All methods are tested using 39,644 characters in three differen t fonts. (C) 1997 Pattern Recognition Society. Published by Elsevier S cience Ltd.