OCR Corpus is mainly a sequence that has the recognition technique can use of HMMs for the text images coupled with the first experiment in the image Database best autoresponder I We trained on fax data present the characters from UW Database I to developed from 2.2% to 1.7% - a reduction by using HMMs for the four fonts, the above problem of language model. Since in speech recognition (CSR) techniques (4.2) that require any form the table to reliably adapted model, we decided to other features not presented sources estimated inherently with some minority fonts. The improve recognizing frames) and computer fonts. We used a training set from other characters in the real corpus to the benefits of copying machines, copying and Yarman-Vural and faxed images. For our experiment, we trained on the system to Chinese data, we would be possible transitions of character.