Comparison of the Multi Layer Perceptron and the Nearest Neighbor Classifier for Handwritten Numeral Recognition

The work presents the results of an investigation conducted to compare the performances of the Multi Layer Perceptron (MLP) and the Nearest Neighbor (NN) classifier for handwritten numeral recognition problem. The comparison is drawn in terms of the recognition performance and the computational requirements of the individual classifiers. The results show that a two-layer perceptron performs comparably to a NN like standard pattern classifier in recognizing unconstrained handwritten numerals, while being computationally more cost effective. The work signifies the usefulness of the MLP as a standard pattern classifier for recognition of handwritten numerals with a large feature set of 96 features.