Anti-spoofing for contactless fingerprint readers

1. An anti-spoofing method for a contactless fingerprint reader, the method comprising steps of:

receiving, from said contactless fingerprint reader, first images of a subject successively captured with and without flash light;

computing pixel differences between the image captured with flash light and the image captured without flash light to estimate a foreground area in these images as a shape of said subject;

extracting a feature vector of said shape to classify it into a positive class or a negative class based on machine learning;

calculating a value representing a texture pattern of an image of said subject captured with flash light and received from said contactless fingerprint reader to classify said value into a positive class or a negative class based on machine learning;

receiving, from said contactless fingerprint reader, second images of said subject captured from different angles;

computing from said second images a depth map of said subject to classify said depth map into a positive class or a negative class based on machine learning;

receiving, from said contactless fingerprint reader, one or more spatial-temporal signals representing a change of a distance between said subject and one or more locations of said contactless fingerprint reader;

extracting feature vectors of said one or more spatial-temporal signals to classify said feature vectors into a positive class or a negative class based on machine learning by:

decomposing each signal by a discrete wavelet transform to calculate an energy of approximation coefficients and an energy of detailed coefficients; and

dividing a sub-band energy from said energies by a sum of said energies to calculate a relative energy of the signal as said feature vector of the signal; and

detecting a spoofing attack if there is at least one negative classification result from the above classifying steps.

2. The method according to claim 1 , said step of calculating a value representing a texture pattern comprising steps of:

defining a circle by a predetermined central pixel and a predetermined radius on said image captured with flash light;

for each neighboring pixel on said circle, calculating a pixel difference between said neighboring pixel and said predetermined central pixel, comparing said pixel difference with a predetermined threshold to obtain a result 1, 0, or −1;

calculating a sum of said results for all the neighboring pixels on said circle and multiplying said sum by two to the power of a total number of the neighboring pixels.

3. A non-transient computer memory comprising:

a computer program with instructions which, when the program is executed by a processor of an anti-spoofing device, cause the device to carry out a method having the steps:

receiving, from said contactless fingerprint reader, first images of a subject successively captured with and without flash light;

computing pixel differences between the image captured with flash light and the image captured without flash light to estimate a foreground area in these images as a shape of said subject;

extracting a feature vector of said shape to classify it into a positive class or a negative class based on machine learning;

calculating a value representing a texture pattern of an image of said subject captured with flash light and received from said contactless fingerprint reader to classify said value into a positive class or a negative class based on machine learning;

receiving, from said contactless fingerprint reader, second images of said subject captured from different angles;

computing from said second images a depth map of said subject to classify said depth map into a positive class or a negative class based on machine learning;

receiving, from said contactless fingerprint reader, one or more spatial-temporal signals representing a change of a distance between said subject and one or more locations of said contactless fingerprint reader;

extracting feature vectors of said one or more spatial-temporal signals to classify said feature vectors into a positive class or a negative class based on machine learning by:

decomposing each signal by a discrete wavelet transform to calculate an energy of approximation coefficients and an energy of detailed coefficients; and

dividing a sub-band energy from said energies by a sum of said energies to calculate a relative energy of the signal as said feature vector of the signal; and

detecting a spoofing attack if there is at least one negative classification result from the above classifying steps.

4. The non-transient computer memory of claim 3 , wherein

wherein the instructions for causing the processor to perform the step of calculating a value representing a texture pattern comprises instructions to cause the processor to perform the steps:

defining a circle by a predetermined central pixel and a predetermined radius on said image captured with flash light;

for each neighboring pixel on said circle, calculating a pixel difference between said neighboring pixel and said predetermined central pixel, comparing said pixel difference with a predetermined threshold to obtain a result 1, 0, or −1;

calculating a sum of said results for all the neighboring pixels on said circle and multiplying said sum by two to the power of a total number of the neighboring pixels.

5. An anti-spoofing device comprising:

a memory;

a processor which is adapted to:

receive, from a contactless fingerprint reader, first images of a subject successively captured with and without flash light;

compute pixel differences between the image captured with flash light and the image captured without flash light to estimate a foreground area in these images as a shape of said subject;

extract a feature vector of said shape to classify it into a positive class or a negative class based on machine learning;

calculate a value representing a texture pattern of an image of said subject captured with flash light and received from said contactless fingerprint reader to classify said value into a positive class or a negative class based on machine learning;

receive, from said contactless fingerprint reader, second images of said subject captured from different angles;

compute from said second images a depth map of said subject to classify said depth map into a positive class or a negative class based on machine learning;

receive, from said contactless fingerprint reader, one or more spatial-temporal signals representing a change of a distance between said subject and one or more locations of said contactless fingerprint reader;

extract feature vectors of said one or more spatial-temporal signals to classify said feature vectors into a positive class or a negative class based on machine learning by:

decomposing each signal by a discrete wavelet transform to calculate an energy of approximation coefficients and an energy of detailed coefficients; and

dividing a sub-band energy from said energies by a sum of said energies to calculate a relative energy of the signal as said feature vector of the signal; and

detect a spoofing attack if there is at least one negative classification result from the above classifying steps.

6. The anti-spoofing device of claim 5 , wherein

to calculate a value representing a texture pattern the anti-spoofing device is adapted to:

define a circle by a predetermined central pixel and a predetermined radius on said image captured with flash light;

for each neighboring pixel on said circle, calculate a pixel difference between said neighboring pixel and said predetermined central pixel, comparing said pixel difference with a predetermined threshold to obtain a result 1, 0, or −1;

calculate a sum of said results for all the neighboring pixels on said circle and multiplying said sum by two to the power of a total number of the neighboring pixels.

7. An anti-spoofing system comprising:

a contactless fingerprint reader having:

one or more light sources adapted to emit light under different lighting conditions;

a first camera adapted to successively capture first images of a subject captured with and without flash light;

at least one second camera adapted to capture second images from different angles;

one or more sensors adapted to generate one or more spatial-temporal signals representing a change of a distance between said subject and one or more locations of said contactless fingerprint reader; and

a communication module for sending said first images, said second images, and said one or more spatial-temporal signals to an anti-spoofing device; and

an anti-spoofing device having:

a memory;

a processor which is adapted to:

receive, from a contactless fingerprint reader, first images of a subject successively captured with and without flash light;

compute pixel differences between the image captured with flash light and the image captured without flash light to estimate a foreground area in these images as a shape of said subject;

extract a feature vector of said shape to classify it into a positive class or a negative class based on machine learning;

calculate a value representing a texture pattern of an image of said subject captured with flash light and received from said contactless fingerprint reader to classify said value into a positive class or a negative class based on machine learning;

receive, from said contactless fingerprint reader, second images of said subject captured from different angles;

compute from said second images a depth map of said subject to classify said depth map into a positive class or a negative class based on machine learning;

receive, from said contactless fingerprint reader, one or more spatial-temporal signals representing a change of a distance between said subject and one or more locations of said contactless fingerprint reader;

extract feature vectors of said one or more spatial-temporal signals to classify said feature vectors into a positive class or a negative class based on machine learning by:

decomposing each signal by a discrete wavelet transform to calculate an energy of approximation coefficients and an energy of detailed coefficients; and

dividing a sub-band energy from said energies by a sum of said energies to calculate a relative energy of the signal as said feature vector of the signal; and

detect a spoofing attack if there is at least one negative classification result from the above classifying steps.

8. The anti-spoofing system according to claim 7 , wherein said light sources are provided with films optimizing flash light reflection and enhancing texture patterns of human skin.

9. The anti-spoofing system according to claim 7 , wherein said sensors are ToF sensors respectively provided on outer regions of a surface of the contactless fingerprint reader and surrounding the cameras arranged in inner regions of the surface.

10. The anti-spoofing system according to claim 7 , wherein said at least one second camera is a stereo camera having at least two lenses adapted to capture said second images respectively.

11. The anti-spoofing system according to claim 7 , wherein said at least one second camera are two cameras adapted to capture said second images respectively.

12. The anti-spoofing system of claim 7 , wherein the processor is adapted to calculate the texture pattern by:

defining a circle by a predetermined central pixel and a predetermined radius on said image captured with flash light;

for each neighboring pixel on said circle, calculating a pixel difference between said neighboring pixel and said predetermined central pixel, comparing said pixel difference with a predetermined threshold to obtain a result 1, 0, or −1;

calculating a sum of said results for all the neighboring pixels on said circle and multiplying said sum by two to the power of a total number of the neighboring pixels.