Signature verification is among the oldest biometric identification methods, and the automatization of its process is a significant field of research. The off-line approach, which promises the widest field of applicability makes its decision solely based on the scanned images of the signatures: but because of the limited information content there is still much room for improvement in this field.
In my paper I study a solution which mimics the approach of forensic handwriting experts, by identifying, matching and comparing smaller features of the signatures (such as the length or number of baselines, or the area of loops) in a group of original signatures in order to decide if the examined writing is original or not. Although the matching of these features is an easy task for a human it is much harder for a computer to tackle. Therefore in my work I created an algorithm, which can solve this multidimensional, weighted problem. After its realisation, by the analyzation of the results we can make observations that may contribute to the constant improvement of feature matching in the field of signature verification.