Deep learning-based fingerprint feature extraction and classification techniques

Abstract

Fingerprint recognition that is most commonly used in biometrics, has been installed in many security facilities as well as in smartphones that we use frequently, and the range of fingerprint recognition is gradually expanding. It is important to improve the performance of fingerprint recognition technology itself, and it is also important to prepare countermeasures against security vulnerabilities. In this dissertation, we propose a feature extraction technique for fingerprint recognition and a fake fingerprint detection technique.

First, it is a pore extraction technique. As technological developments have enabled high-quality fingerprint scanning, sweat pores, one of the Level 3 features of fingerprints, have been successfully used in Automatic Fingerprint Recognition Systems (AFRS). Since the pore extraction process is a critical step for AFRS, high accuracy is required. However, it is difficult to extract the pore correctly because the pore shape depends on the person, region, and pore type. To solve the problem, we propose a pore extraction method using deep convolutional neural networks (DCNN) and pore intensity refinement. The deep networks are useful to employ to detect fine holes using a wide area of the fingerprint image. The deep networks are used to detect pores in detail using a large area of a fingerprint image. We generate a pore map using the deep networks, then refine the pore information by finding local maxima to identify pores with different intensities in the fingerprint image. The experimental results show that our pore extraction method achieves state-of-the-art performance.

Second, it is a fake fingerprint detection technique. Recently, the importance of fake fingerprint detection technique has been growing as fingerprint recognition technology is used in many places of daily life including a smartphone. In order to detect fake fingerprints, we propose a fingerprint liveness detection method using an enhanced fingerprint image, transfer learning, and multi-task learning. The proposed method uses multi-channel input images consisting of a fingerprint image and an enhanced fingerprint image with emphasis on shape information, fine-tune the pre-trained model, and perform multiple classification tasks simultaneously in training stage. The enhanced fingerprint image has the effect of emphasizing the shape difference between the live fingerprint image and the fake fingerprint image. Fine-tuning the existing model is useful when the training dataset is small. The purpose of multi-task learning is to improve the accuracy of fake fingerprint classification by simultaneously learning fake fingerprint classification and more complex spoof material classification. The proposed method shows state-of-the-art performance and has high detection accuracy for fake fingerprint images made with unknown spoof materials not used for training.