Recognition of off-line hand printed English Characters, Numerals and Special Symbols

摘要

The generic process of Optical Character Recognition (OCR), an area of intensive research in the field of Artificial Intelligence, Pattern Recognition and Computer Vision, aims to recognize text from scanned document images, where data can be in machine printed or hand written format. Optical Character Recognition can improve the interaction between man and machine in various applications including data entry, office automation, digital library, banking applications, health insurance and tax forms etc. Much of work has been done in the recognition of machine printed characters in various languages with considerably good efficiencies, however making robust recognition engines that can be put to recognize hand written and hand printed data with commendable recognition rates still remains as an active area of research owing to the challenges like diverse human handwriting style, variation in shape, angle and style of characters. Taking into account the challenges and scope for improvement in this domain, the work of off-line character recognition of hand printed document images containing English Characters-Uppercase and Lowercase, Numerals and Special Characters has been presented. Statistical, Geometric and Directional Feature Extraction techniques have been applied over segmented character image. Classification was done using Multilayer perception neural network (NN) with back propagation and Support vector machine (SVM) classifier. The recognition rates achieved were up to 98% for Numerals, 96.5% for Special Characters, 95.35% for Uppercase Characters, 92% for Lowercase Characters. The system for combined data set-Characters, Numerals and Special Symbols resulted out to be 92.167% accurate, using SVM as classifier.