An Efficient Method for Devanagari Character Recognition
Balaji R. Bombade, Dharmpal Dronacharya Doye, Uday V. Kulkarni · International journal of mathematics and computation · 2015
Handwritten character recognition is always been a challenging task for the researchers. Algorithms for feature extraction introduced till date are not up to snuff to recognize the Indian handwritten scripts productively. This paper describes two novel methods proposed for feature extraction of Indian handwritten scripts. The first algorithm utilises Hough Transform and Directional Chain Code (HTDCC) and consists of two stages to extract lines and curves from handwritten characters. HTDCC collects the lines and curves as features from handwritten character objects. The subsequent approach partitions character image in blocks and extracts Block Based Structural Features (BBSF) from respective blocks. Unlike HTDCC, it is used to collect joints between lines and curves as features of character image. Finally, Support Vector Machine (SVM) is used for classification. The proposed techniques yield 90.34 and 94.21 percent accuracy on two standard Devnagari character databases.