Exploring Image Watermarking in Digital Collections of Ancient Coins

This paper investigates invisible blind watermark approaches as a security mechanism to track ownership in online digital collection of ancient coin images. Recent deep-based architectures are compared in addition to traditional transformation-domain approaches. Besides evaluating watermarking for the task-specific images, this analysis takes into account challenging attacks obtained by combining both geometric or photometric attacks to test the actual effectiveness of compared methods for practical purposes. Moreover, the sequential application of pair of watermarking methods is considered with the intention of increasing the robustness of the joint watermarking and as an additional form of attack. Experimental evaluation shows that deep-based watermarking is robust to most of the base attacks and that sequential joint watermarking can improve upon the robustness of the original methods. Finally, although the strength of recent approaches to common attacks is sufficient, complex combined attacks remain still a challenge.