Authentication of random patterns by finding a match in an image database

摘要

This paper describes an approach for finding the exact match of a random texture pattern in an image database. It is assumed that the random pattern may be misregistered relative to its representation in the database and it may have missing parts. The specific application of the proposed algorithm is authentication of legal documents, bank notes or credit cards, where thin fibers are embedded randomly into the document medium during medium fabrication. Random patterns guarantee extremely high uniqueness, i.e. low probability of two documents being identical, and thus offer very effective anti-counterfeiting measures. Image matching is achieved by a three step hierarchical procedure, where each step uses a separate database. The preliminary tests of the authentication algorithm included experiments with over 10,000 synthetic images and 50 real images. In all cases the algorithm has given correct results. This paper details the algorithm, considerations on document uniqueness, and testing procedures.