Synthetic Fingerprint Generation for Biometric Systems Using Attention-Based GANs

Abstract

Biometric systems have become an essential part of secure authentication frameworks due to their reliability, uniqueness, and resistance to replication. However, the development and evaluation of these systems are often constrained by limited and imbalanced fingerprint datasets. Synthetic fingerprint image generation offers a viable solution to this challenge by enhancing data diversity and volume. This research presents an innovative method for high-quality fingerprint synthesis using Attention-Based Generative Adversarial Networks (GANs). The proposed model integrates attention mechanisms within the GAN architecture to improve the spatial consistency, texture realism, and minutiae preservation of the generated images. Through extensive experimentation, the synthesized fingerprints were validated against benchmark datasets using metrics such as Fréchet Inception Distance (FID), Structural Similarity Index (SSIM), and minutiae matching accuracy. The results indicate that the attention-enhanced GAN significantly outperforms baseline GANs and traditional augmentation approaches. This advancement not only facilitates the development of robust biometric models but also contributes to privacy-preserving research by minimizing the need for real biometric data.