Fingerphoto Deblurring Using Attention-guided Multi-stage GAN

Using fingerphoto images acquired from mobile cameras, low-quality sensors, or crime scenes, it has become a challenge for automated identification systems to verify the identity due various acquisition distortions. A significant type of photometric distortion that notably reduces quality is blurring image. This paper proposes deep deblurring model restore ridge information degraded by image blurring. As core our model, we utilize conditional Generative Adversarial Network (cGAN) learn distribution natural patterns. We perform several modifications enhance reconstructed (deblurred) fingerphotos proposed model. First, develop multi-stage GAN in coarse-to-fine framework. framework enables maintain consistency process at different resolutions. Second, propose guided attention module helps generator focus mainly on blurred regions. Third, incorporate verifier as an auxiliary adaptive loss function force preserve ID during process. Finally, evaluate effectiveness through extensive experiments multiple public datasets well real-world fingerphotos. In particular, method achieves 5.2 dB, 8.7%, and 7.6% improvement PSNR, AUC, EER, respectively, compared state-of-the-art method.