Please use this identifier to cite or link to this item: http://localhost:8081/xmlui/handle/123456789/2968
Authors: Kumar, Nitin
Issue Date: 2012
Abstract: In this work an attempt is made to develop "handwritten English character recognition system". The report describes the process of character recognition using the hybrid algorithm of Back Propagation and Genetic Algorithm for the recognition of uppercase alphabets. In the survey, it is found that back propagation is although an efficient technique for training multilayer feed forward network. But it suffers from scaling, local minima like problems. And also it is studied that Genetic Algorithm is good optimization technique. It is effective for global search of large, poorly understood spaces. The system is thrown through numbers of steps of character recognition system like preprocessing, segmentation and feature extraction. Distinctive features for each character are extracted. Those features are passed to hybrid algorithm of Back Propagation neural network and Genetic Algorithm. With the increasing growth of internet, also the demand of online information system has increased. There are various applications like in post offices; banks etc. where character recognition systems are used. In this theses report three methods for Roman handwritten Character recognition_ system are used and there result are analysis in this chapter. In first method, structural features and binary intensity of Roman handwritten Character is used. In this method, first Roman handwritten Character is divided into 81 segment parts. Then chain code direction feature and binary intensity is calculated for every part. Thus there are total 450 (300 + 150) features, _ one segment. part of Roman handwritten Character. Thus no. of features is .very high, so to reduce the features, all features in one segment of Roman handwritten Character image is added.-. - Thus it reduces the no. of features to 100 features. However even 100 feature for one Roman handwritten Character is very high for neural network. So these features are further reduces to 50 per Roman handwritten Character. This is done by taking top 50 Roman handwritten Character of a Roman handwritten Character image. Now with these 50 features per Roman, handwritten Character is trained by ANN and results are obtained. _ In second method, some global features with moment invariant vectors are used to identify a Roman handwritten Character. In this method 8 global properties like height to width, ratio, max vertical and horizontal projections, vertical and horizontal centers, image area, max vertical and horizontal projection points are used. Seven moment invariant vectors are calculated. Thus there are 15 features per Roman handwritten Character used. Now these features are used to train ANN.
Other Identifiers: M.Tech
Research Supervisor/ Guide: Mukherjee, Shaktidev
metadata.dc.type: M.Tech Dessertation
Appears in Collections:MASTERS' DISSERTATIONS (Electrical Engg)

Files in This Item:
File Description SizeFormat 
EEDG22057.pdf7.19 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.