Two Methods for Recognition of Hand Written Farsi Characters

摘要

Optical character recognition (OCR) is one of the active bases of sample detection topics. The current study focuses on automatic detection and recognition of hand written Farsi characters. For this purpose; we proposed two different methods based on neural networks and a special post processing approach to improve recognition rate of Farsi uppercase letters. In the first method, we extracted wavelet features from borders of character images and learned a neural network based these patterns. In the second method, we divided input characters into five groups according to the number of their components and used a set of appropriate moment features in each group and classified characters by the Bayesian rule. In a post-processing stage, some structural and statistical features were employed by a decision tree classifier to reduce the misrecognition rate. Our experimental results show suitable recognition rate for both methods.