Multiple Classifiers for Unconstrained Offline Handwritten Numeral Recognition

In this work we propose an approach that includes use of multiple classifiers for unconstrained handwritten numeral recognition. The objective of the present work is to provide efficient and reliable techniques for recognition of handwritten numerals. Features used for classification of numerals are directional features. The classifiers used to solve the complex problem of digit recognition are multi-layer perceptron (MLP) classifier, Combinations of learning vector quantization (LVQ) classifier and K nearest neighbor (KNN) classifier. Outputs from these classifiers are further combined by using a discriminant function. Experiments and results show that the present method is robust for recognizing handwritten numerals.