Neural Networks and Gradient-Based Learning in OCR

A large proportion of today's commercial Optical Character Recognition systems (OCR) and Handwriting Recognition Systems (HWR) use neural networks at the core of the recognition engine. Comparisons on standard databases show that Neural Networks, particularly multi-layer networks, offer a good combination of speed, generality, simplicity, and flexibility. They are also particularly well-suited for the large input dimension required for shape recognition tasks such as character recognition.