The Use of Synthetic Finger Vein Images in Deep Learning Pre-training
Boyang Yu, Peiyu Fang · 2022
Finger vein biometrics has been extensively applied to personal verification. In order to extract and classify features, a famous pertained model VGG-16 was used. In addition, due to lack of availability of large-scale finger vein databases, this paper implements a virtual digital vein image generation algorithm using biodynamics based on previous methods. A virtual data set is generated and is used for the pre-training of classification model. The result is that the pre-trained model using virtual datasets can achieve faster convergence and higher accuracy. Thus, the similarity between the synthesized image and the real image is proved, and the generated dataset is proved to be available in the classification model pre-training task.