However, it has been suggested that it can be trained by using
character groups, such as words.
Arabic OCR is widely regarded as a hard problem, and is presently
being approached with Hidden Markov Models or with brute force
learning of large data sets. The commercial software can be quite
expensive.