A refinement of stroke structure for printed Thai character recognition

In this paper, it presents a very efficient recognition system for printed Thai characters. The major techniques are using structural features of stroke and setting production rules for classification. For the former, a technique is employed to re-order the character image components derived from skeleton of character image as based on Thai scripts and glyphs. In the latter, a set of production rules is established with a stroke patterns for character clustering and further for character classification. In this work, 12 clusters for Thai consonants, digits and special characters are grouped together, including 8 clusters for Thai vowel signs and tone marks. The proposed system is tested with over 1,000,000 Thai characters of multi-fonts and multi-sizes, the result shows that it could yield a very high accuracy rate in those tests of recognition.