An efficient Devanagari character classification in printed and handwritten documents using SVM

摘要

With the increased demand, exploration and globalization of digitized Devanagari documents, many printed and handwritten mono-lingual character recognition techniques have evolved since last two decades. This paper presents an efficient Devanagari character classification model using SVM for printed and handwritten mono-lingual Hindi, Sanskrit and Marathi documents, which first preprocesses the image, segments it through projection profiles, removes shirorekha, extracts features, and then classifies the shirorekha-less characters into pre-defined character categories. The experiments performed on proposed system obtained average classification accuracies of 99.54% and 98.35% for printed and handwritten images, respectively, and showed better performance than other techniques.