Off-line Signature Verification Using Multidirectional Run-Length Features

Abstract

Run-length features have shown its effectivity in off-line signature verification. Configurations of these handcrafted features include four direction-based features. In this paper, we propose to add further spatial information of the signature to the standard run-length features. Such information is worked out in two stages: firstly, beyond classical four directions, more directions are studied. Secondly, improving the knowledge of each direction by combining the information of the neighbor directions. This new configuration has been used in two classifiers, one based on Euclidean distance and another based on a one-class support vector machine. We outperformed previous performances with the proposed configuration regarding the last ICFHR 2018 competition on Thai student signatures and GPDS-960 signature database.