Improving Digital Library Support for Historic Newspaper Collections

DVD-ROM Appendix available with the print copy of this thesis.National and international initiatives are underway around the globe to digitise the vast treasure troves of historical artefacts they contain and make them available as digital libraries (DLs). The developed DLs are often constructed from facsimile pages with pre-existing metadata, such as historic newspapers stored on microfiche or generated from the non-destructive scanning of precious manuscripts. Access to the source documents is therefore limited to methods constructed from the metadata. Other projects look to introduce full-text indexing through the application of off-the-shelf commercial Optical Character Recognition (OCR) software. While this has greater potential for the end user experience over the metadata-only versions, the approach currently taken is best effort in the time available rather than a process informed by detailed analysis of the issues. In this thesis, we investigate if a richer level of support and service can be achieved by more closely integrating image processing techniques with DL software. The thesis presents a variety of experiments, implemented within the recently published open-source OCR System (Ocropus). In particular, existing segmentation algorithms are compared against our own based on Hough Transform, using our own created corpus gathered from different major online digital historic newspaper archives