Toward comprehensive and effective palmprint reconstruction attack.

The challenge posed by the template-based reconstruction attack significantly impacts the security and privacy of biometric systems. Current reconstruction techniques rely on extensive training data or encounter limitations in adaptability, resulting in subpar reconstruction performance. In this paper, we propose a black-box palmprint template reconstruction method based on the modified Progressive GAN (ProGAN), which achieves a substantial success rate in attacking deep-learning-based and hand-crafted-based templates. Our approach incorporates the dropout mechanism into the generator of ProGAN and introduces a Double Reuse Training Strategy to enable effective training of the reconstruction network despite limited data. Furthermore, we devise a novel Scale-Adaptive Multi-Texture Complementarity loss, enhancing the texture quality of reconstructed images. We conduct extensive experiments on diverse palmprint recognition techniques. The resulting reconstructed images exhibit exceptional image quality. Additionally, we thoroughly examine the security and privacy aspects of the palmprint recognition algorithm based on the insights gained from the reconstruction attacks. [Display omitted] • The first palmprint reconstruction method suitable for both code and deep templates. • Dropout improves ProGAN architecture for reconstruction task. • Two strategies, namely SAMTC and DRTS, enhance reconstruction performance. • A comprehensive assessment system for reconstructed image quality. • This work reveals deep-based methods are more resistant to reconstruction attack. [ABSTRACT FROM AUTHOR]