Differentiating between live and spoof fingers in fingerprint analysis by machine learning

1. A method performed in a fingerprint analysis system for liveness detection, the method comprising:

acquiring a plurality of time-sequences of images, each of the time-sequences showing a respective finger as the respective finger engages a detection surface of a fingerprint sensor during a time period when the respective finger is approaching and being pressed against the detection surface until the finger covers a whole detection area of the detection surface, the detection area corresponding to an area imaged in the time-sequences, each of the time-sequences comprising at least a first image and a last image showing a fingerprint topography of the respective finger, wherein the respective fingers of a first set of the time-sequences are known to be live fingers and the respective fingers of a second set of the time-sequences are known to be spoof fingers; and

training a machine learning algorithm on the plurality of time-sequences to produce a model of the machine learning algorithm for differentiating between a live finger and a spoof finger,

wherein, for each of the time-sequences, acquiring of the time-sequence has been triggered when the detection surface indicates a presence of the respective finger but the respective finger is not yet in contact with the detection surface over the whole detection area of said detection surface.

2. The method of claim 1 , wherein the fingerprint analysis system comprises the fingerprint sensor, and wherein the plurality of time-sequences are acquired by means of said fingerprint sensor.

3. The method of claim 1 , wherein the plurality of time-sequences are acquired from a database.

4. The method of claim 1 , wherein the plurality of time-sequences are acquired in real-time from respective electronic devices in association with authentication operations performed thereon.

5. The method of claim 1 , wherein each of the time-sequences has been obtained at a rate of at least 20 images per second.

6. The method claim 1 , wherein the machine learning algorithm is a Support Vector Machine (SVM) algorithm, a neural networks algorithm, a deep learning algorithm, or Multi-Layer Perception (MLP) algorithm.

7. The method claim 1 , wherein the machine learning algorithm comprises, for each time-sequence, forming a one-dimensional vector of pixel information from the time-sequence of images.

8. The method of claim 7 , wherein Principal Component Analysis (PCA) is applied to the one-dimensional vector to reduce the dimensionality thereof.

9. The method of claim 8 , wherein one or more Support Vector Machines are applied to the one-dimensional vector with reduced dimensionality.

10. The method of claim 1 , wherein the machine learning algorithm comprises, for each image of each time-sequence, detecting edges of the image to form an edge image.

11. The method of claim 10 , wherein the machine learning algorithm further comprises superpositioning the edge images of the time-sequences to form an accumulated edge image per time-sequence.

12. The method of claim 10 , wherein the machine learning algorithm further comprises extracting histograms representing the accumulated edge image.

13. The method of claim 12 , wherein one or more Support Vector Machines are applied to a vector of said histograms.

14. A computer program product comprising a non-transitory computer readable medium storing computer-executable components for causing a fingerprint analysis system to perform the method of claim 1 when the computer-executable components are run on processing circuitry comprised in the fingerprint analysis system.

15. A method performed in an electronic device comprising a fingerprint sensor of a fingerprint analysis system in association with an authentication operation performed by the electronic device and the fingerprint sensor, the method comprising:

acquiring a time-sequence of images of a candidate finger as the candidate finger engages a detection surface of the fingerprint sensor during a time period when the candidate finger is approaching and being pressed against the detection surface until the candidate finger covers a whole detection area of the detection surface, the detection area corresponding to an area imaged in the time-sequence, wherein the time-sequence comprises at least a first image and a last image of the time-sequence showing a fingerprint topography of the candidate finger,

applying a model of a machine learning algorithm to the time-sequence; and

based on a result of the applied model, determining whether the candidate finger is a live finger or a spoof finger,

wherein the acquiring is triggered when the detection surface indicates a presence of the candidate finger but the candidate finger is not yet in contact with the detection surface over the whole detection area of said detection surface.

16. The method of claim 15 , wherein the determining comprises determining that the candidate finger is a live finger, whereby the method further comprises:

granting access to functionality of the device based on said determining.

17. The method of claim 15 , wherein the model is preprogrammed in the electronic device.

18. The method of claim 15 , wherein the time-sequence is obtained at a rate of at least 20 images per second.

19. A fingerprint analysis system comprising:

processing circuitry; and

data storage storing instructions executable by said processing circuitry whereby said fingerprint analysis system is operative to:

acquire a plurality of time-sequences of images, each of the time-sequences showing a respective finger as the respective finger engages a detection surface of a fingerprint sensor during a time period when the respective finger is approaching and being pressed against the detection surface until the finger covers a whole detection area of the detection surface, the detection area corresponding to an area imaged in the time-sequences, each of the time-sequences comprising at least a first image and a last image showing a fingerprint topography of the respective finger, wherein the respective fingers of a first set of the time-sequences are known to be live fingers and the respective fingers of a second set of the time-sequences are known to be spoof fingers; and

train a machine learning algorithm with the plurality of time-sequences to produce a model of the machine learning algorithm for differentiating between a live finger and a spoof finger,

wherein, for each of the time-sequences, acquiring of the time-sequence has been triggered when the detection surface indicates a presence of the respective finger but the respective finger is not yet in contact with the detection surface over the whole detection area of said detection surface.

20. The fingerprint analysis system of claim 19 , further comprising the fingerprint sensor.

21. An electronic device comprising:

a fingerprint sensor;

processing circuitry; and

data storage storing instructions executable by said processing circuitry whereby said electronic device is operative to:

acquire a time-sequence of images of a candidate finger as it engages a detection surface of the fingerprint sensor during a time period when the candidate finger is approaching and being pressed against the detection surface until the candidate finger covers a whole detection area of the detection surface, the detection area corresponding to an area imaged in the time-sequence, wherein the time-sequence comprises at least a first image and a last image showing a fingerprint topography of the candidate finger;

apply a model of a machine learning algorithm to the time-sequence; and

based on a result of the applied model, determine whether the candidate finger is a live finger or a spoof finger,

wherein the acquiring is triggered when the detection surface indicates a presence of the candidate finger but the candidate finger is not yet in contact with the detection surface over the whole detection area of said detection surface.