Global Approach for Script Identification using Wavelet Packet Based Features

In a multi script environment, an archive of documents having the text regions printed in different scripts is in practice. For automatic processing of such documents through Optical Character Recognition (OCR), it is necessary to identify different script regions of the document. In this paper, a novel texture-based approach is presented to identify the script type of the collection of documents printed in seven scripts, to categorize them for further processing. The South Indian documents printed in the seven scripts - Kannada, Tamil, Telugu, Malayalam, Urdu, Hindi and English are considered here The document images are decomposed through the Wavelet Packet Decomposition using the Haar basis function up to level two. Gray level co-occurrence matrix is constructed for the coefficient sub bands of the wavelet transform. The Haralick texture features are extracted from the co-occurrence matrix and then used in the identification of the script of a machine printed document. Experimentation conducted involved 2100 text images for learning and 1400 text images for testing. Script classification performance is analyzed using the K-nearest neighbor classifier. The average success rate is found to be 99.68%.