Image based retrieval and keyword spotting in documents

Abstract

The attempt to move towards paperless offices has led to the digitization of large quantities of printed documents for storage in image databases. Thanks to advances in computer and network technology, it is possible to generate and transmit huge amount of document images efficiently. An ensuing and pressing issue is then to find ways and means to provide highly reliable and efficient retrieval functionality over these document images from a vast variety of information sources. Optical Character Recognition (OCR) is one powerful tool to achieve retrieval tasks, but nowadays there is a debate over the trade-off between OCR-based and OCR-free retrieval, because of OCR errors and wastage of time to OCR the entire collection into text format. Instead, image-based retrieval using document image similarity measure is a much more economical alternative. Till now, many methods have been proposed to achieve different sub-tasks, all of which contribute to the final retrieval performance. This chapter will present different methods for presenting word images and preprocessing steps before similarity measure or training and testing and discuss different algorithms or models for achieving keyword spotting and document image retrieval.