Online recognition of handwritten Telugu script characters

The hand-held devices revolutionized the way users interact and demands support for regional languages. Handwritten interfaces simplify the communication in regional languages without the need for multiple key presses. The handwriting interfaces need to recognize characters while writing. Thus the proposed work attempted to design an efficient SVM-based recognizer for online Recognition of isolated Telugu handwritten characters. The Legendre-Sobolev series coefficients for each of the X and Y coordinates with degree 12 augmented with the sine and cosine angles of the first-to-third and first-to-last vectors are used as features. These 28 simple and easy to extract features combined with number of strokes based pre-classification strategy reported a recognition rate of 90%. The proposed methodology achieved comparable recognition performances with only 28 features as against to hundreds of features used in literature. It is suitable for online recognition as it does not require preprocessing and computation of features may be overlapped with the writing of the character.