Arabic Word Spotting Based on Key-Points Features

Abstract

Abstract Word spotting is an efficient alternative to OCR (Optical Character Recognition) for understanding historical manuscripts by matching features of words. However, due to the large variability in the Arabic script, the retrieving results are still not satisfactory. In this work a novel method based on key-points features is proposed. Key-points features can capture both topological and local characteristics. Feature vectors were extracted from key-points and a distance function was proposed for word matching. To reduce space matching, the connected components (CCs) were clustered into metaclasses in a soft manner using a Gaussian mixture model, then a query CC was spotted by matching it only with CCs which belong to its meta-class. The experiments were carried out on a benchmark consisting of Arabic historical manuscripts by IBEN SINA and shows promising results.