Recognition of Printed Text Based on Hidden Markov Model

Abstract

Automatic recognition of printed text is of high importance in modern IT applications. Recognition of text for lateen scripted language is readily in use for a long time. For cursive script languages (such as Arabic language) recognition of text is not available as a robust one with a reliable performance. More improvements still exist to reduce average of incorrect words, rather then no constraints on the limit of words of a specific language. Numerous approaches were tried in recognition of text but recognition of Arabic text based on Hidden Markov model seems to be the most promising one because of its ability to discriminate cursive scripts. This paper provides an off-line system to recognize printed Arabic text by using hidden Markov model with the aid of the algorithm that segment the text lines into connected parts then into characters. By looking on the results given by the designed recognition system it is found that a recognition rate (94.9 %) can be achieved. Such rate is in the same order of rates of recognition researches viewed in previous studies. This rate can still be improved. The language used in building the system is Matlab V7.6 (R2008a).