Comprehensive analysis of least significant bit and difference expansion watermarking algorithms for online signature signals

This paper investigates the efficacy of non-reversible and reversible watermarking techniques applied to online signature signals through an analysis of Least Significant Bit and difference expansion algorithms. The impact of watermark insertion is assessed using dynamic time warping recognition accuracies on a large database. Our experimental section comprises identification rates and verification errors for both random and skilled forgeries, involving 330 users and 25 skilled forgeries per user. Notably, our findings reveal that online signature signals can withstand up to 7 bits per sample insertion in the least significant bit algorithm, with minimal reduction in recognition accuracies. Similar robustness is observed with the difference expansion algorithm. This resilience is noted when employing z-score normalization and an expanded feature set, encompassing delta and delta-delta parameters. The presented results underscore the robustness of the proposed watermarking methods in preserving the integrity of online signature recognition capabilities. Watermarking is currently proposed to differentiate between human-produced signals and those generated by artificial intelligence, with applications including replay attack detection. This study explores the potential application of this technique to online handwriting signals without compromising biometric recognition accuracy. The proposed watermarking techniques not only preserve biometric recognition integrity but also provide a promising solution to distinguish human-generated signals from AI-generated counterparts, a growing challenge in the age of synthetic media.