Method of extracting features from a fingerprint represented by an input image

1. A method for extracting features of interest from a fingerprint represented by an input image, the method being characterized in that it comprises the implementation, by data processing means ( 21 ) of a client equipment ( 2 ), of steps of:

(a) Direct estimation from the input image of at least one candidate angular deviation of an orientation of said input image with respect to a reference orientation, by means of a convolutional neural network (CNN);

(b) Recalibration of said input image as a function of said estimated candidate angular deviation, so that the orientation of the recalibrated image matches said reference orientation;

(c) Processing said recalibrated image so as to extract said features of interest from the fingerprint represented by said input image.

2. A method according to claim 1 , wherein step (a) comprises the identification of at least one potential angular deviation class of orientation of said input image with respect to the reference orientation among a plurality of predetermined potential angular deviation classes, each potential angular deviation class being associated with an angular deviation value representative of the class, the estimated candidate angular deviation(s) having as values the representative value(s) of the identified class(es).

3. A method according to claim 2 , wherein each class is defined by an interval of angular deviation values.

4. A method according to claim 3 , wherein said intervals are continuous and form a partition of a given range of potential angular orientation deviations.

5. A method according to claim 4 , wherein said intervals are all of the same size.

6. A method according to claim 2 , wherein step (a) comprises determining an orientation vector of said input image associating each of said plurality of potential angular deviation classes to a score representative of the probability that said input data belongs to said potential angular deviation class.

7. The method according to claim 2 , comprising a prior training step (a0), by data processing means ( 11 ) of a server ( 1 ), from a fingerprint image database where each image is already associated with a class of said plurality of predetermined potential angular deviation classes, from parameters of said CNN.

8. A method according to claim 7 , wherein said training uses a Sigmoid-type loss function.

9. A method according to claim 1 , wherein a recalibrated image is generated in step (b) for each estimated candidate angular deviation, step (c) being implemented for each recalibrated image.

10. A method according to claim 1 , wherein said CNN is of the residual network type.

11. A method according to claim 1 , wherein step (c) is also carried out on the non-recalibrated input image.

12. The method according to claim 1 , wherein said features of interest to be extracted from the fingerprint represented by said input image comprise the position and/or orientation of minutia.

13. The method according to claim 1 , wherein said fingerprint represented by the input image is that of an individual, the method further comprising a step (d) of identifying or authenticating said individual by comparison of features of interest extracted from the fingerprint represented by said input image, with the features of reference fingerprints.

14. A method according to claim 1 , wherein said CNN does not perform any object detection task in the input image.

15. A non-transitory computer program product comprising code instructions for the execution of a method according to claim 1 of extraction of features of interest from a fingerprint represented by an input image, when said program is executed on a computer.

16. A non-transitory computer readable storage means readable by a computer equipment on which a computer program product comprises code instructions for the execution of a method according to claim 1 of extraction of features of interest from a fingerprint represented by an input image.