Handwritten indic script identification from document images—a statistical comparison of different attribute selection techniques in multi-classifier environment

Abstract

Script identification from document images is an essential task before choosing script-specific OCR for a Multi-lingual/Multi-script country like India. The problem becomes more complex when handwritten document images are considered. Several techniques have been developed so far for HSI (Handwritten Script Identification) problem and the work is still in progress. But the issue of dimensionality reduction of the feature set for script identification problem has not been addressed in the literature till date. This paper presents a statistical performance analysis of different attribute selection techniques in a multi-classifier environment for HSI problem on Indic scripts. A GAS (Greedy Attribute Selection) technique for HSI problem has also been proposed here. Encouraging outcomes are found observing the complexities of handwritten Indic scripts.