Classification of Japanese Kanji using principal component analysis as a preprocessor to an artificial neural network

Abstract

The authors applied principal component analysis (PCA) to the problem of classifying handwritten Kanji characters. PCA is a statistical tool which can yield substantial data reduction by representing each pattern in terms of a relatively small subset of orthonormal features (principal components) extracted from the input set. A PCA preprocessor to an artificial neural network has been used to reduce the dimensionality of a set of handwritten Kanji patterns to less than 5% of that of the original images. Reconstructions of the patterns from the preprocessed versions are quite impressive. Preliminary results yield nearly 90% correct classification of exemplars of 40 different Kanji characters, and also indicate that reconstruction requires more information than classification. These results demonstrate the effectiveness of PCA as a preprocessor for neural networks.