Nandinagari Palm Leaf Word Image Retrieval System

This paper provides the first attempt for the recognition of Nandinagari handwritten word in a handwritten Palm Leaf manuscript. We take a set of very aged Palm Leaf and take a representative sample containing over 1000 characters and perform a set of preprocessing steps including background subtraction, de noising using Gaussian filter, contrast enhancement using histogram equalization and binarization using adaptive thresholding technique to obtain high quality readable manuscript. The words are subsequently extracted using annotation method to get a set of 100 meaningful Handwritten Nandinagari words of different sizes. A dictionary of these vocabulary words is formed and stored. Effective corner based feature extraction technique is applied to these images and the corresponding scale and rotation invariant features are extracted and stored in the database. The query word image is then compared with the dictionary words and the matched similar word images are retrieved.