Mahsa Vafaie, Sven Hertling

With the increasing amount of digitised documents, automatic document analysis is gaining more and more attention. Despite all the progress in the field, historical documents still pose challenges to document analysis systems. They differ from modern documents in many aspects, such as digitisation format, paper quality, degradation, font type, and amount of noise.
An information retrieval pipeline can benefit greatly from the detection of stamps, especially in the context of digital cultural heritage, as stamps usually contain information relating to creation, distribution, and storage of the documents. The goal of this thesis is to detect stamps on historical documents using a combination of computer vision and machine learning techniques and algorithms.