Hierarchical optical character recognition system design based on the Hopfield neural networks

Abstract

Pattern recognition is a scientific discipline dealing with the methods for object description and classification and the Optical Character Recognition (OCR) is one of its fields of research. In this paper a hierarchical optical character system design is presented. Classification strategy based on the Hopfield neural networks and image processing methods are described. The characters for recognition are Cyrillic alphabet capital letters. The first step in the design is a neural network testing with the real scanned document in order to see how the network works. Based on the results of testing with one Hopfield neural network, some common sources of error in this system were found. These sources of error were a base for new improvements in the system. Next step is, therefore, the addition of new binary image processing parameters and new pre-processing and post-processing techniques for a typical error's elimination. After testing the same real scanned document again, the obtained results showed that this new and improved system decreased an error probability significantly.

Keywords

• pattern recognition
• character recognition
• optical recognition
• pattern recognition system
• hierarchy system
• neural network
• associative memory