Neural networks in off-line text recognition: A review

Abstract

Artificial neural networks (ANN) have helped optical character recognition (OCR) systems to achieve the stat of the art performance by providing powerful tools for the classification and probability estimation as well as their efficient computational architecture which facilitates hardware implementation. It is found that most commercial OCR systems employ certain forms of ANNs. The narrowing of the performance gap between OCR systems and human operators on unconstrained handwritten text and cursive scripts is described.