G.A. Yuryev, L.S. Kuravsky

The technology intended for recognition of text information by visually impaired persons is under consideration. It is intended for transformation of initial text images obtained online with the aid of a video camera from usual books, computer monitors and similar sources to their audio representation. Initial images after their acquisition are converted into corresponding contour shapes by a proper wavelet transform and, then, decomposed into character strings, which are recognized by means of three different techniques, one of which combines capabilities of wavelet transforms and relaxation neural networks. Important practical advantage of this technique is the use of minimal character training pattern sets. Recognized character strings are then spoken by a speech synthesizer. Features of the implemented hardware-software system are given.