Tamil Numerals Identification

Abstract

This paper describes a system to identify Tamil numerals using a two-stage approach, for a subset of Tamil alphabet or numerals. In the first stage, five sets of each Tamil numeral are trained and in the second stage un-trained Tamil numerals are tested against the trained numerals. We use some pre-processing steps to enhance the images. Resizing has been performed on the images for having a standard size for all the images. Gray-scale conversion has been performed on the images for reducing dimensionality of images, which are required for further processing. Binarization has been performed using a threshold technique to change the pixel value to 0 or 1. After binarization feature extractions of the images are performed as rows, columns signal variations. Training and testing processes are carried out to evaluate the feature extraction.