Paper-Based Watermark Extraction with Image Processing

This thesis presents frameworks for the digitisation, localisation, extraction and graphical representation of paper-based watermark designs embedded in paper texture. There is a growing need for this among librarians and antiquarians to aid with identification, wider accessibility, and providing a further level of document imaging for preservation. The proposed approaches are designed to handle manuscripts with interference such as recto and verso writing, and defects such as non-uniform paper structure, physical damage, etc.

A back-lighting scanning technique is used for capturing images of paper, followed by a selection of intelligent image processing operations, rather than alternatives such as radioactive techniques. This technique requires low cost equipment, and produces a fast and safe solution to capturing all details on paper, including watermarks, and laid and chain lines patterns.

Two approaches are presented: the first takes a bottom-up approach and deploys image processing operations to enhance, filter, and extract the watermark, and convert it into a graphical representation. These operations determine a suitable configuration of parameters to allow optimal content processing, in addition to the detection and extraction of chain lines. The second approach uses a model of the back-lighting effect to locate a watermark in pages of archaic documents. It removes recto information, and highlights remaining ‘hidden’ data, and then presents a statistical approach to locate watermarks from a known lexicon.

Work is further presented on reconstructing features of the paper mould by aggregating the success of the foregoing steps: this permits an analysis of ‘twin’ watermarks.

Results are presented from comprehensively scanned eighteenth and nineteenth century manuscripts, including two unusual copies of the Quran, an Islamic Prayer, and various historical documents.