RETRIEVAL FROM HISTORIC AMHARIC DOCUMENT IMAGES

Abstract

ABSTRACT The world has experienced a phenomenal growth of the size of multimedia data and especially document images, which have been increased thanks to the ease to create such images using scanners or digital cameras. Thus, huge quantities of document images are created and stored. Digitalization enables us to better preserve knowledge so that it passes from generation to generation. This is essential since we are now in a place where knowledge is power. On top of this digitalization improves the organization of document, thus reduce space the physical needed to store and the time spent to find and use the document. But this is not an easy task. Document image retrieval is a method which mainly focuses on retrieving document image from document image database. There are two approaches to carry out document image retrieval: recognition based and recognition free retrieval. Several works are done to develop Amharic document image retrieval (DIR) system. But it fails to investigate historical document images. On the top of that these works focus on the level of noise available. This study proposed a way to improve the DIR system by integrating a noise removal technique for historical Amharic document images and it investigates the type of noise exists on historical Amharic document images. Six morphological, from basics to complex, noise filtering operations and two thresholding technique namely, Otsu global thresholding method and Sauvola local thresholding technique, are investigated and step wise combination of open-close with Sauvola thresholding algorithm outperform any other noise filtering and thresholding techniques combination. The selected noise removal scheme is finally integrated with Amharic DIR system for evaluation purpose. Accordingly, system performance shows that it registers on the average 91.67%, 68.86 and 76.89 recall, precision and F-measure respectively. The