Offline Signature Verification Based on Bag-of-Visual Words Model Using KAZE Features and Weighting Schemes

The familiar use of handwritten signatures in various applications (e.g., credit card authentication) increases the need for automated verification methods. However, there is still room for improvement in the performance of automated systems under various writing conditions compared to human beings, especially forensic document examiners (FDEs). Furthermore, even with modern techniques, obtaining as much information as possible from the limited samples available remains challenging task. Therefore, further research is required to improve the performance of automated systems. In this study, to improve the performance of offline signature verification, a new approach based on a bag-of-visual words (BoVW) model is adopted. The novelty features of the proposed approach are following: 1) considering the cognitive processing of visual information by FDEs to improve the performance of offline signature verification, 2) using an approach based on the BoVW model to implement the FDEs' cognitive process for feature extraction, 3) incorporating weighting schemes based on term frequency-inverse document frequency to enhance the discriminative power of each visual word, 4) adopting KAZE features in the BoVW model to consider the contour information of strokes more effectively, and 5) detecting the KAZE features in both the strokes and background space to introduce not only the stroke itself but also the various relations between strokes. The promising performance of the proposed approach is shown by using an evaluation method with a popular CEDAR signature dataset.