AN ANALYSIS OF NEURAL NETWORK CLASSIFIER FOR HANDWRITTEN CHARACTER RECOGNITION

Abstract

The recognition of handwritten numerals or characters has applications in the field of pattern recognition, document processing and analysis. However, the recognition of handwritten numerals or characters is a challenging task for machine because the writing styles vary from person to person. Handwritten numeral or characters recognition is in general a benchmark problem of Pattern Recognition and Artificial Intelligence. Compared to the problem of scanned numeral recognition, the problem of handwritten numeral recognition is compounded due to variations in shapes and sizes of handwritten English numerals. This makes the recognition of handwritten numerals or characters an intensive area of research in pattern recognition. Considering all these, the problem of handwritten numeral recognition is addressed under the present work in respect to handwritten English numerals. This dissertation presents a new approach for handwritten numeral recognition method using the MLP (Multi Layer Perceptron) neural network for classification purpose. To achieve the desired task of recognition, a numeral is first enclosed inside a bounding box whose size is normalized to 42 x 30 pixels. The numeral bounding box is partitioned into 9 sub boxes and features are extracted using bounding box approach is given as input to the MLP classifier. Many images with different fonts and styles have been processed in MATLAB environment using the aforementioned classifier. It is observed from the results that the proposed technique successfully classifies handwritten numerals used in our experiment. The proposed scheme proves its utilization for recognition handwritten English numerals represented with the said feature set. This approach of recognition of handwritten English numerals by MPL classifier can also be extended to include handwritten characters of English alphabet.