 |
University of Maryland, College Park, Maryland USA and Centre for Strategic Infocomm Technologies, Singapore |
 |
|
University of Maryland, College Park, Maryland USA and Centre for Strategic Infocomm Technologies, Singapore |
|
The challenges related to the analysis of large heterogeneous collections of document images ultimately encompass almost all aspects of the field of document image processing. The written language takes on many forms that differ in presentation and content, yet the trained individual can interpret the visual language rather simply. The goal of document analysis is ultimately to be able to make an informed interpretation about the intended message of the visual language.
In this project, we will develop specific modules of interest to the sponsors related to Triage, Enhancement, Segmentation, and Content Labeling. The work will be accomplished by researchers in the Laboratory for Language and Media Processing (LAMP) at the University of Maryland and integrated with an existing infrastructure for document image analysis. The proposal contains an in-depth discussion of the problems and lays a roadmap for addressing them. We anticipate further conversations with the sponsors will focus the directions outlined here.
We assume, at the lowest level, we are given an image that may contain useful document related content. Our goal is to first determine if the image does contain document content, then to enhance and process it to the point of sufficient layout metadata to support down stream content processing such as optical character recognition. To support focused research we will develop the necessary tools, gather ground truth, visualize results, and provide efficient implementations of the algorithms we develop.