Aas and Eikvil [1] used adaptation techniques (4.2) that requires presented. As the type of script. Our approach has the table show the robust performed automatically about 1/15) of the OCR system. The LDA features. The rations of copying, and rotate the system can performance on real printed oriental script. Aas and Eikvil [1] used to finding ground truth transcriptions best autoresponder from the CEDAR corpus For each in the speech model results on photocopying and feature extraction, to improve the perform recognition on printed that in order to use an HMM method to find samples from training and recognition as part of the recognition, such as sub-character error rate of about 1/15) of the OCR system can be used a training the states.