Design of a Decision Tree to Classify Similar Looking Characters using Subimages for Kannada Script

Kannada script has large number of vowels, consonants, conjuncts and combination of these in inflectional and agglutinative manner. Moreover, Kannada characters have higher similarity in shape and higher variability across fonts, making recognition of characters a difficult task. Even though many Kannada characters have similar appearance, they differ in small image region. These unique small regions or subimages can be utilized to discriminate between the characters. In this research, a decision tree is effectively built by grouping similar looking characters and partitioning them into categories based on these subimages. Experiments have been conducted on Kannada database of size 3360 and an overall classification rate of 93.8 % is observed