LAUSR

Handwritten character recognition has been a challenging task and it finds uses in many real time scenarios such as language translation, automatic Braille transliteration, automatic cheque book reading, automatic scanning of handwritten forms etc. This paper takes two Indian scripts namely Devanagari and Odia for handwritten character recognition and compares different gradient based features such as LBP, LDP, HOG and LOOP which can be used for learning the character pattern by a support vector classifier on various parameters. The paper uses two existing databases of handwritten Devanagari and Odia characters to train the support vector classifiers and compares the results of various features selection. It proposes the best possible feature for Devanagari and Odia character recognition based on graphical comparisons of parameters such as accuracy, training time and recognition rate. The maximum accuracy achieved on the Devanagari dataset of 92000 characters is 95.65% and the maximum accuracy achieved on the relatively small Odia dataset of 15400 characters is approximately 99%. The paper further investigates for the Devanagari characters getting misclassified more frequently.