DSB-GAN: Generation of deep learning based synthetic biometric data

Abstract

Deep learning-based generative networks have brought a significant change in the generation of synthetic biometric data. Synthetic biometric data finds applications in developing biometric systems and testing them on a large amount of data to analyze their performance on extreme load scenarios or run simulation for health care personnel training. Generally, biometric datasets have fewer training samples, due to which deep learning models do not train well. In the proposed DSB-GAN, a generative model based on convolutional autoencoder (CAE) and generative adversarial network (GAN) is used to generate realistic synthetic biometrics for various modalities such as fingerprint, iris, and palmprint. This generated data ensures the availability of data that is not available in general due to various undesired factors like distortion and corruption of data. The model is resource efficient and generates diverse biometric samples as compared to state-of-the-art methods. © 2022 Elsevier B.V.