A scene text-based image retrieval system

Abstract

The rapidly increasing popularity of digital images raise a question of how to automatically index and help users navigate such libraries of images. One approach is to extract text appearing in images which often gives an indication of a scene's semantic content. However, it can be challenging since the text is often embedded in a complex background. This paper presents a novel approach for querying images using scene text. The text in natural scene images is localized and extracted properly by a novel mechanism. Then, an indexing and retrieval scheme is introduced. The proposed methods are evaluated on the public database of the ICDAR 2003 competition. Experimental results are very encouraging and suggest that these algorithms can be used in image retrieval applications.