Analysis on Efficient Handwritten Document Recognition Technique Using Feature Extraction and Back Propagation Neural Network Approaches

Abstract

Today, Handwritten recognition becomes a very much thrust area in the field of pattern recognition and image processing. Handwritten recognition methods are used in real-life fields such as banking checks, car plates number identification, recognition of ZIP code, mail sorting, reading of different commercial forms, etc. The work is presented in this paper a system of handwriting of English document recognition based on feature extraction of the character. Almost 400 handwritten numerals are collected from different datasets as for sample for the classification purposes. The main work is presented here are consisting of several steps like preprocessing, feature extraction and Multi-Layer Perceptron model of neural network for classifying handwritten digits separately. Basically, Offline Recognition of English handwriting using multilayer perceptron network or back propagation networks are described throughout the entire work. First, the English alphabets are used as features introducing features sets of different English handwritten documents. Then the neural network is used tos train the datasets. Different types of training methods of back propagation network are used for calculating performance of the entire system. The recognition methods are based on back propagation network for analyzing the classification performance of handwritten documents. Here the system achieves the accuracy more than 90% using this efficient back propagation neural network based classification and feature extraction methods using morphological operations based zones separation scheme of digits. Here, the performance performance parameter like sensitivity, specificity, recall, accuracy provides more than 90% of rates indicates that better classification of handwritten documents.