Recognition of Identical Shape Handwritten Devnagari Vowels

Devnagari is most popular script in India. There has been a significant improvement in the research related to the recognition of handwritten Devanagari characters in the past few years, but accurate recognition is a difficult task due to variations in shapes of the same character with different writer and identical shape characters. Most of the identical shape Devnagari characters are in vowels as compared to consonants. This paper proposes a novel method of recognizing identical shape handwritten Devnagari vowels. Recognition is carried out with two stage classifiers and multiple feature extraction methods. Support Vector Machine (SVM) based approach is suggested for pre-classification. For this pre-classifier chain code histogram features are used. Then vowels misclassified or confused with its identical vowel are found and grouped together. Each group vowels are applied to set of feature extraction methods specifically foreground pixel distribution, Intersection/junction features, chain code histogram features, and zone density features. These features are applied individually to second stage classifiers. Artificial neural network and SVM are used as second stage classifiers. Finally second stage classifiers outputs are combined with weighted majority voting scheme, for final decision. This approach of multi-stage classification improves the recognition rate of identical shape vowels to 94.61%.