Fingerprint authentication based on deep convolutional descent inversion tomography

Fingerprint authentication is widely used in various areas. While existing methods effectively extract and match fingerprint features, they encounter difficulties in detecting wet fingers and identifying false minutiae. In this paper, a fast fingerprint inversion and authentication method based on Lamb waves is developed by integrating deep learning and multi-scale fusion. This method speeds up the inversion performance through deep fast inversion tomography (DeepFIT) and uses Mask R-CNN to improve authentication accuracy. DeepFIT utilizes fully connected and convolutional operations to approach the descent gradient, enhancing the efficiency of ultrasonic array reconstruction. This suppresses artifacts and accelerates sub-millimeter-level fingerprint minutia inversion. By identifying the overall morphological relationships of various minutia in fingerprints, meaningful minutia representing individual identities are extracted by the Mask R-CNN method. It segments and matches multi-scale fingerprint features, improving the reliability of authentication results. Results indicate that the proposed method has high accuracy, robustness, and speed, optimizing the entire fingerprint authentication process.