Synergy of foreground–background images for feature extraction: Offline signature verification using Fisher vector with fused KAZE features. (July 2018)

• Record Type: Journal Article
• Authors: Okawa, Manabu
• Is Part Of: Pattern recognition. Volume 79(2018:Jul.)
• Page Start: 480
• Page End: 489
• Publication Date: 2018-07
• Subjects: Biometrics -- Forensics -- Signature verification -- Fisher vector -- KAZE features -- Fusion strategy -- Support vector machine; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2018.02.027
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: KAZE features from foreground and background signature images show good performance. Fused KAZE features with representation-level fusion further improve performance. FV provides a more precise spatial distribution of the characteristics per writer. PCA for the FV provides a more compact vector without significant performance loss. This method yields lower error rates than existing signature verification systems. Abstract: Offline signature verification has been accepted as a tool for individual authentication. To address the remaining challenges and improve the discriminative power, this study proposes a new feature extraction approach based on a Fisher vector (FV) with fused KAZE features detected from both foreground and background signature images using a recent fusion strategy. Experimental results demonstrate the following: (1) KAZE features from foreground and background signature images show good performance, respectively; (2) fused KAZE features from foreground and background signature images improve performance; (3) adoption of the FV provides a more precise spatial distribution of the characteristics per writer; (4) while an FV with representation-level fusion produces a high-dimensional vector, principal component analysis for the original FV can provide a more dimensionally compact vector without significant performance loss; (5) with the popular MCYT-75 signature dataset, the proposed method yields significantly lower error rates than existing state-of-the-art offline signature verification methods.