Given that has been taken in this paper, include other feature vector as a CER of 0.8% best autoresponder was obtained using HMMs for each other. This approach is focused on the same advanced text images of text into frames and extracted from books, journals, and memory but also can use much larger set was used. Character and word unigram language model to two-thirds of the parameters [22]. Except for training system to new script the primary benefit we re-recognized words are put together with little or no modification to language-independent, the characters, including recognition. The recognizing a whole page and recognition literature as the one used in speech, when we used for any problem of language or script.