A robust braille recognition system

Abstract

Braille is the most effective means of written communication between visually-impaired and sighted people. This paper describes a new system that recognizes Braille characters in scanned Braille document pages. Unlike most other approaches, an inexpensive flatbed scanner is used and the system requires minimal interaction with the user. A unique feature of this system is the use of context at different levels (from the pre-processing of the image through to the post-processing of the recognition results) to enhance robustness and, consequently, recognition results. Braille dots composing characters are identified on both single and double-sided documents of average quality with over 99% accuracy, while Braille characters are also correctly recognised in over 99% of documents of average quality (in both single and double-sided documents). © Springer-Verlag 2004.