Notice of Violation of IEEE Publication Principles - An pproach to recognize handwritten bengali numerals for postal automation

Abstract

A recognition approach for handwritten Bengali numerals and its application for Bangladesh Postal system are presented in this paper. The approach consists of preprocessing, feature extraction and recognition. Using the canny edge detector the top left corner of the post-code box has been detected and the four handwritten numeral images have been segmented. Each image then goes through the steps of normalization, filtering and thinning. Kirsch mask has been used to get the edges through the horizontal, vertical, right and left diagonal. Principal component analysis (PCA) has been used for dimension reduction as the final feature vector consists of the four directional feature vector along with the normalized image. Then the output of the PCA is passed to a trained Support Vector Machine (SVM) to determine which class the input belongs to. Experiments demonstrate that the average recognition rate, error rate and reliability achieved by the proposed system are 92.5%, 7.5% and 92.5% respectively.