Framework for Automatic Identification of Paper Watermarks with Chain Codes

Abstract

In this dissertation, I present a new framework for automated description, archiving, and identification of paper watermarks found in historical documents and manuscripts. The early manufacturers of paper have introduced the embedding of identifying marks and patterns as a sign of a distinct origin and perhaps as a signature of quality. Thousands of watermarks have been studied, classified, and archived. Most of the classification categories are based on image similarity and are searchable based on a set of defined contextual descriptors. The novel method presented here is for automatic classification, identification (matching) and retrieval of watermark images based on chain code descriptors (CC). The approach for generation of unique CC includes a novel image preprocessing method to provide a solution for rotation and scale invariant representation of watermarks. The unique codes are truly reversible, providing high ratio lossless compression, fast searching, and image matching. The development of a novel distance measure for CC comparison is also presented. Examples for the complete process are given using the recently acquired watermarks digitized with hyper-spectral imaging of Summa Theologica, the work of Antonino Pierozzi (1389 – 1459). The performance of the algorithm on large datasets is demonstrated using watermarks datasets from well-known library catalogue collections.