Abstract

This paper presents an alternative technique for “signature verification”. The technique employs template matching for feature extraction and Rules 3-ext inductive learning algorithm to extract the necessary set of rules and to verify a signature. 15 of 3 × 3 masks were used to represent a signature. Each signature (or pattern) is presented by the frequencies of the masks used. The system was trained using 144 signatures (16 signatures belonging to 9 different persons each). The system has been tested using many unseen signatures and the ability to correctly classify them was found to be 97%.

Key words: Inductive learning, signature verification, machine learning, image processing.