TITLE: An Efficient Realization of An OCR System Using HDL

AUTHORS: Azam Beg

PUBLICATION/VENUE: International Conference on Artificial Intelligence (ICAI'08), Jul 2008, pp. 74-78.

ABSTRACT:

This paper presents a Verilog model of an artificial neural network for Arabic character recognition. A neural network by nature is a non-linear system that presents some unusual challenges when realized in digital domain. A main feature of our proposed model is that it does not require hardware-intensive resources, such as dividers and multipliers. A character recognition accuracy of 80.3% was achieved with the model. The flexible nature of the model allows experimentation with any other different set of neural network weights, without affecting the overall network structure. The network model can be easily adapted to other similarly-written Middle- Eastern/Asian languages, such as Persian (Iran), Urdu (Pakistan), Pushto (Afghanistan), etc. Potential applications of such a system are in portable devices such as pen-shaped text readers/recognizers; Braille-aware or low-vision text-to-speech devices; or in large-scale character recognition systems such as postal mail systems, bank-check processing, etc.

DOWNLOAD: PDF

CITATIONS: Google | Yahoo