Project Title

Information Access from Document Images of Indian languages

Web Abstract

Development content aware image processing algorithms for robust and efficient recognition and retrieval from Indian language document images is proposed. Our image processing algorithms aim at improving the quality of document images by removing the noise and low resolution artifacts by adopting content aware shape-based morphological filters. A set of recognizers will be built using state of the art machine learning techniques such as deep learning for handwritten, typewritten and low resolution document images where the existing technologies are insufficient. For hard and noisy handwritten documents, we propose holistic keyword spotting techniques to reduce search space and complement the recognition based approaches. We will also build and demonstrate information access and retrieval schemes over a joint space of image features and noisy text, so as to enable a set of immediate practical applications. The methods will be validated on two different focussed collections during the project.