Holistic recognition of handwritten Tamil words

Abstract

In this paper, we describe a writer independent method for recognizing offline unconstrained handwritten Tamil words/strings. Touching and overlapping are main bottleneck of handwriting recognition and to overcome this, we adopted a holistic approach here. To handle various handwritings of different individuals, at first, some pieces of preprocessing work are done on an input word. Next, Gabor based features are computed on the processed word. These Gabor features along with other geometric features of the word image are then fed to an SVM classifier for recognition. In our experimental study, we have used 4270 samples (collected from the class of 217 country names) written in Tamil and obtained 86.36% recognition rate.