Word-wise South Indian Script Identification using GLCM and Radon Features

Abstract

This paper presents a hybrid features for identification of south Indian scripts in word-wise and it has used three classifiers. We have used two kinds of features namely Radon and Gray Level Co-occurrence Matrix (GLCM) and combination of Radon and GLCM features. For identification purpose LDA, KNN and SVM classifiers have been employed. For the experiment proposed work considered the 6 languages scripts; Roman, Devnagari, Kannada, Telugu, Tamil and Malayalam. This proposed work considered the Word Image Dataset for 11 Languages form MILE Lab IISC in this dataset proposed work considered 6 languages with 5000 for each scripts, this makes total of 30,000 word images. We have made the total of five bi-lingual combinations of south Indian scripts. To extract features; GLCM and Radon Features are considered (4 features of GLCM, 11 features, for Radon we obtained 98.80% from KNN for the Roman and Kannada combination, for GLCM 88.20% obtained by SVM for the Roman and Kannada from SVM Classifier and from combination of Radon and GLCM we have obtained the accuracy of 98.90% for Roman and Kannada combination scripts.