Printed Thai character recognition using fuzzy-rough sets

This paper proposes the method of fuzzy rough sets for the recognition of Thai Characters. We divide the classification process into 2 levels, coarse and fine classification. Both levels of classification have the same processes, applying the rough set lower approximation and then using fuzzy rough sets. The difference between the two levels is in the features of the input data used for classifying. There are 40 coarse groups and some of them need not pass through the second level of classification. We trained this system with 2816 training samples, which were composed of 4 fonts and 4 sizes of characters. The system is tested with an unknown sample, which is composed of 7 fonts and 7 sizes of characters; 4 fonts and 4 sizes of the training sample are inclusive. The accuracy of this proposed system is as high as 89%.