Neural network based optical character recognition system

Abstract

This paper presents an application of a neural network in the optical character recognition (OCR) system. It introduces general architecture of modern OCR systems, discussing each module in detail. Specific contribution of this paper is novelty of the character extraction and segmentation, by considering them as image features. MSER (Maximally Stable Extremal Regions) feature detector is applied, discussing numerical and practical restrictions for character segmentation and recognition. The neural network is trained for character recognition and applied on the appropriate example.