DeepFingerPCANet: Automatic Fingerprint Classification Using Deep Learning

Fingerprints are expanding in popularity, and the fingerprint datasets are becoming increasingly huge; they are recorded using a range of sensors embedded in smart devices like mobile phones and personal computers. The difficulty of fingerprint recognition systems is worsened when they are obtained using different sensors, which is one of the main challenges. Fingerprints can be categorized in a database to reduce the search space and speed up the query response. However, classifying cross-sensor fingerprints is a challenging problem. An efficient and robust solution is to use a convolutional neural network (CNN), but designing its architecture is time-consuming. In order to automatically design a CNN model for fingerprint classification, we developed a strategy that uses pyramidal clustering, principal component analysis (PCA), and the ratio of the between-class scatter to within-class scatter to determine the number of filters and the number of layers in the model automatically. It aids in the building of lightweight CNN models that are efficient and speed up fingerprint classification. We validated the proposed method on two benchmark datasets, FingerPass and FVC2004, which feature noisy, low-quality fingerprints obtained via live scan devices and various sensors. Compared to existing fingerprint classification methods and well-known pre-trained models, the newly developed models perform noticeably better.