A Complete Scheme of Word Spotting System for the Balinese Palm Leaf Manuscripts

Abstract

The word spotting system is urgently needed to be able to work with thousand pages of digitized ancient Balinese palm leaf manuscript. This system will be very helpful for the scholars to perform a keyword searching in all manuscript collections with a query image. In this paper, we present a complete scheme of word spotting system for the Balinese palm leaf manuscripts. Our proposed complete scheme of word spotting system for the Balinese palm leaf manuscripts consists of three main sub schemes: the offline patch images extraction process, the feature extraction method, and the patch ranking scheme. We applied six filters of Gabor from six different orientations for the patch image extraction process, and we proposed the use of 64 Gabor filters which are combined with seven different Zoning methods for the feature extraction methods. We also propose an adaptive sliding patch algorithm to spot all possible patches in manuscript page and the patch ranking scheme to rank the spotting results. For the test and evaluation, we used a published dataset of AMADI_LontarSet. Our scheme shows a very promising results. The word spotting system is able to spot many query patch images in different manuscript pages.