English Character Recognition Using Robust Back Propagation Neural Network

Abstract

OCR deals with the handwritten or printed character recognition with the help of digital computers and soft computing. The scanned images of characters and numbers are used as input for system which is analyzed and transformed it into character codes, normally in ASCII format, which is taken for the data processing. Presently, there are lots of issues in recognition of characters and numbers, which can degrade the performance of the system in various ways. Mainly, the rate of recognition is not improved due to distributed neighborhood pixels of an image. Also, there are some techniques used for the OCR are having lack of contrast levels which is well known by fading of the image. So in this paper, the most important concern is to take such measures to enhance the performance of the system for automatic recognition of characters. Here, the operations are performed on the handwritten English alphabets. The dataset is selected from the Chars74K with different shapes and preprocessed it which deals with filtering and edge detection. Then, in the feature extraction process, features are extracted by using independent component analysis and swarm intelligence is used for feature vector selection. Classification of images are done with the back propagation neural network which gives an effective learning approach. The precise contribution which is evaluated in this research work is the uniqueness of classifications using a combination of the feature extraction and feature optimization (instance selection) using extraction of feature vectors. The performance of the developed system is measured in terms of recognition rate, sensitivity and specificity compared with the benchmark.