AlexNet Convolutional Neural Network architecture with cosine and hamming similarity/distance measures for fingerprint biometric matching

In information security, fingerprint verification is one of the most common recent approaches for verifying human identity through a distinctive pattern. The verification process works by comparing a pair of fingerprint templates and identifying the similarity/matching among them. Several research studies have utilized different techniques for the matching process such as fuzzy vault and image filtering approaches. Yet, these approaches are still suffering from the imprecise articulation of the biometrics’ interesting patterns. The emergence of deep learning architectures such as the Convolutional Neural Network (CNN) has been extensively used for image processing and object detection tasks and showed an outstanding performance compared to traditional image filtering techniques. This paper aimed to utilize a specific CNN architecture known as AlexNet for the fingerprint-matching task. Using such an architecture, this study has extracted the significant features of the fingerprint image, generated a key based on such a biometric feature of the image, and stored it in a reference database. Then, using Cosine similarity and Hamming Distance measures, the testing fingerprints have been matched with a reference. Using the FVC2002 database, the proposed method showed a False Acceptance Rate (FAR) of 2.09% and a False Rejection Rate (FRR) of 2.81%. Comparing these results against other studies that utilized traditional approaches such as the Fuzzy Vault has demonstrated the efficacy of CNN in terms of fingerprint matching. It is also emphasizing the usefulness of using Cosine similarity and Hamming Distance in terms of matching.