Versatile search of scanned Arabic handwriting

Abstract

Searching handwritten documents is a relatively unexplored frontier for documents in any language. Traditional approaches use either image-based or text-based techniques. This paper describes a framework for versatile search where the query can be either text or image, and the retrieval method fuses text and image retrieval methods. A UNICODE and an image query are maintained throughout the search, with the results being combined by a neural network. Preliminary results show positive results that can be further improved by refining the component pieces of the framework (text transcription and image search). © 2008 Springer-Verlag Berlin Heidelberg.