Method for extracting a signature of a fingerprint and device implementing said method
A method for extracting a signature of a fingerprint shown on a source image is described. For this purpose, the source image is transformed into the frequency domain. One or more ridge frequencies are next determined by means of a convolutional neural network applied to said transformed image, n being an integer greater than or equal to 1. The source image is normalized in response to said n ridge frequencies determined. One or more signatures of said fingerprint are finally extracted from said n normalized images.
1. A method for extracting a signature of a fingerprint shown on a source image comprising:
transforming said source image into the frequency domain;
applying a trained convolutional neural network to said transformed image, said neural network determining as an output for each ridge frequency in a set of N ridge frequencies a probability that said source image has said ridge frequency as its ridge frequency, N being a positive integer;
selecting the n ridge frequencies associated with the n highest probabilities, wherein n≥2 and n<N;
normalizing said source image in response to each of said n ridge frequencies selected; and
extracting n signatures of said fingerprint from said n normalized images.
2. The method according to claim 1 , wherein transforming said source image into the frequency domain comprises applying a Fourier transform.
3. The method according to claim 1 , wherein the convolutional neural network is of the U-Net type.
4. The method according to claim 1 , wherein extracting n signatures of said fingerprint from said n normalized images comprises extracting characteristics of minutiae belonging to said fingerprint.
5. The method according to claim 4 , wherein said minutiae characteristics comprise the position and/or the orientation of said minutiae.
6. The method according to claim 1 , which further comprises comparing said n extracted signatures with at least one extracted signature of a reference fingerprint associated with an individual and identifying the fact that said fingerprint shown on the source image belongs to said individual in the case where at least one of said n extracted signatures is similar to said at least one extracted signature of said reference fingerprint.
7. The method according to claim 1 , wherein the parameters of said convolutional neural network were learnt from a database of reference fingerprints for which the ridge frequency is known.
8. A non-transitory computer-readable storage medium configured to store instructions for implementing, by a processor, the method according to claim 1 , when said instructions are executed by said processor.
9. A device for extracting a signature of a fingerprint shown on a source image comprising at least one processor configured to:
transform said source image into the frequency domain;
apply a trained convolutional neural network to said transformed image, said neural network determining as an output for each ridge frequency in a set of N ridge frequencies a probability that said source image has said ridge frequency as its ridge frequency, N being a positive integer;
select the n ridge frequencies associated with the n highest probabilities, wherein n≥2 and n<N;
normalize said source image in response to each of said n ridge frequencies selected; and
extract n signatures of said fingerprint from said n normalized images.