Innovative feature sets for machine learning based Telugu character recognition

In this Information age, all sources of information like historic documents, books, manuscripts are digitized and are available all over the world through internet in the form of scanned copies. These scanned images contain valuable information which are available either in colour or black and white for pleasant viewing. Optical Character Recognition (OCR) technology provides facility to search for keywords in these digital copies. In this paper, new method in which building an OCR system for Telugu language script; mainly focussing on the character recognition module. Features extracted through Discrete Wavelet Transform (DWT), Projection Profile (PP) and Singular Value Decomposition (SVD) is evaluated using k-Nearest Neighbour (k-NN) and Support Vector Machine (SVM) classifiers. Most productive results are obtained from the DWT features with SVM classifiers.