Vector of locally aggregated descriptors with KAZE features for offline signature verification

The widespread use of handwritten signatures for idnetity authentication has resulted in a need for automated verification systems that make use of modern electronic devices (e.g., scanners, cameras, and digitizing tablets). However, there is still significant room for improvement in the performance of these automated systems when compared with that of human analysis, particularly that of forensic document examiners (FDEs), under a wide range of conditions. As the consequences of any inaccuracy in these results can cause serious problems, further research is required to improve the performance. In this study, to improve the performance of offline signature verification while reducing calculation costs, a new approach using vector of locally aggregated descriptors (VLAD) is adopted. The novelty features of the proposed approach include the following: 1) the method considers knowledge recently obtained about the cognitive processes engaged by FDEs to improve performance, 2) it incorporates an approach based on VLAD with KAZE features to mimic the FDEs' cognitive processes for feature extraction, and 3) it applies principal component analysis to VLAD to further reduce calculation costs. The promising performance of the proposed approach is demonstrated through its application to a popular CEDAR signature dataset.