Title: Symbolic data analysis-based few-shot learning for offline handwritten signature verification

Authors: Mohamed Anis Djoudjai; Youcef Chibani; Adel Hafiane

Addresses: Laboratoire d'Ingénierie des Systèmes Intelligents et Communicants, Faculty of Electronics and Computer Science, University of Sciences and Technology Houari Boumédiene (USTHB), 32, El Alia, Bab Ezzouar, 16111, Algiers, Algeria ' Laboratoire d'Ingénierie des Systèmes Intelligents et Communicants, Faculty of Electronics and Computer Science, University of Sciences and Technology Houari Boumédiene (USTHB), 32, El Alia, Bab Ezzouar, 16111, Algiers, Algeria ' Laboratoire Pluridisciplinaire d'Ingénierie des Systèmes, Mécanique et Energétique, Institut National des Sciences Appliquées Centre Val de Loire (INSA-CVL), University of Orléans, PRISME, EA 4229, 18022 Bourges, France

Abstract: This paper presents a novel approach for offline handwritten signature verification using few-shot learning and symbolic data analysis. The method effectively handles high intra-class variability and limited data availability, common challenges in signature recognition. The model is trained on dissimilarities from the Signet feature extractor, capturing subtle differences within the same writer's signatures. A new weighted membership function measures similarity between query and reference signatures. The method outperforms traditional approaches, achieving competitive equal error rates on four public datasets (GPDS, CEDAR, MCYT, PUC-PR) using only five genuine reference signatures. The system surpasses state-of-the-art results on GPDS and PUC-PR datasets. This combination of few-shot learning and symbolic data analysis offers robust and efficient signature verification, ideal for real-world applications with scarce labelled data.

Keywords: few-shot learning; FSL; signature verification; intra-class variability; one-class symbolic data analysis classifier; dissimilarities.

DOI: 10.1504/IJBM.2025.145921

International Journal of Biometrics, 2025 Vol.17 No.3, pp.311 - 329

Received: 01 Dec 2023
Accepted: 09 Apr 2024
Published online: 30 Apr 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article