This publication describes techniques and methods implemented on a computing device, directed at detecting latent fingerprint reactivation attacks (e.g., the exploitation of latent fingerprints deposited on fingerprint sensors to gain access to computing devices). In aspects, for each fingerprint authentication event, an on-device fingerprint-matching algorithm extracts small patches from a fingerprint image (a “Verify Image”) and transforms the patches into rotationally-invariant vectors. The fingerprint-matching algorithm then calculates the similarity between the vectors of the Verify Image and vectors of a stored, authentic fingerprint (an “Enrolled Image”). If the computing device validates the fingerprint (e.g., authorizes the fingerprint and determines it is not a latent fingerprint by the steps described below), the computing device permits the user access. The fingerprint image, formerly Verify Image, is now a “Previous Image.” Finally, the fingerprint-matching algorithm merges the vectors from the Enrolled Image and Previous Image using a rotation and translation matrix (a “Previous Matrix”). The computing device temporarily stores the Previous Matrix.
Upon a successive fingerprint authentication event, the algorithm extracts small patches from a new Verify Image, transforms the patches into rotationally-invariant vectors, calculates the rotation and translation matrix relative to the Enrolled Image (a “Verify Matrix”), and computes the similarity between the Previous Matrix and the Verify Matrix. If the rotation and translation matrices are identical, then the computing device can reject the latent fingerprint.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Sammoura, Firas and Bussat, Jean-Marie, "Latent Fingerprint Detection Using Rotationally-Invariant Vectors", Technical Disclosure Commons, (May 19, 2020)