Cross-Sensor Fingerprint Recognition Using Convolutional Neural Network and Canonical Correlation Analysis

The available fingerprint verification systems are based on using one sensor type to capture fingerprints. The difficulty arises when using different sensors for enrolment and query stages; it results in a cross-sensor matching or sensor interoperability problem. However, there is only one available deep learning-based study in the literature that has focused on interoperability between contact-based sensors. There are also a limited number of studies based on deep learning that have addressed the interoperability problem between touch and touchless sensors. This brings the need for more efficient, robust, and automatic methods to solve this problem. This paper proposes the design of an automated deep learning-based method to handle the cross-matching problem. CNN models must be trained using large-scale labelled databases. However, this is a challenge for the cross-sensor fingerprint matching problem as there is no available public database comprising a large number of labelled fingerprints. Hence, a pre-trained CNN is used to extract the features of fingerprints from various sensors. The features extracted from fingerprints for the same finger i captured using different sensor types are not correlated. To ensure that the feature representation is discriminative with the smallest within-class scatter and the largest between-class separation, a Canonical Correlation Analysis (CCA) is applied. It is then followed by calculating the matching score among two fingerprints using the Mahalanobis cosine distance. The proposed method was evaluated on three databases were built. Results reveal that our method outperforms the state-of-the-art technique. In verification mode, the revealed cross EER value for all cases was less than 1, regardless of sensor type. For identification mode, the proposed model achieved a high result for all three databases.