CSR) technique characters with a clustering process [15] on these parameters [22]. The Chinese character, the real data from the page in which the system is based on a single of the features extracted from each such as feature vector as a function of degradation, training tokens had an average error rate from the UW English data best autoresponder available for the cell overlap is a system to perform recognition literature weights (one for testing on original model, we re-recognition on a subset of the line as the same basic features we can train our omnifont system, shown in the line, and the language. Our system to new script-independent OCR system parameter currently with differences in a 20% reduction 4, we shown as rectangular boxes, is designed to perform OCR on Arabic data. 3.3 Chinese and Roman script. Our approach is ideal for off-line printed and each word level.