Recognition of Tamil Palm-Leaf Manuscript Characters using a two-fold classification system

Abstract

This paper reports a system for recognizing Tamil Characters present on Palm-Leaf Manuscripts. A two level classification system consisting of Quadratic SVM and k-NN classifier have been used. On the first level, direction based features of individual character images are extracted and fed to a Quadratic SVM and k-NN classifier. For contradictory outputs a decision is made for next level classification. This second level classification involves a time-frequency visualization of the character image. Further, these are fed to another Quadratic SVM classifier. As a initiative, we have formulated the Tamil Palm-Leaf Character dataset. The proposed system has succeeded in providing satisfactory recognition for Tamil characters extracted from the a set of palm-leaf manuscripts.)

Keywords- Tamil Palm-leaf Manuscript, Character Recognition, SVM, Directional Feature, Wavelet Energy.