Printed gujarati word-matching

Abstract

This report describes an approach to perform word-matching for printed Gujarati script word images. A query word image can be used to match indexed word images of a document for document retrieval. The process starts with binarized, skew-corrected word images on which pre-processing is done to remove noise using median filtering and the images are thinned and segmented into character images. Low-level strokes [2] and high-level strokes [3] are extracted as features from each character image of the query word image. These features are stored in a tree structure to represent each word image. The tree structure of the query word image is compared and matched with tree structure of each word in the printed word image database to identify matches.