Method for detecting fraud by pre-recorded image projection
The invention provides a method for detecting fraud ( 220 ) during biometric authentication of an individual comprising the acquisition ( 210 ), by an optical device, of a sequence of candidate images comprising at least two candidate images of an individual, with view to comparing said sequence with a reference image sequence comprising at least one reference image of an individual acquired during a preliminary enrollment step ( 100 ) by the same optical device, the method comprising the steps of: estimating ( 221 ), from the sequence of candidate images, the geometrical parameters of the optical device having acquired said sequence of candidate images on the individual, and comparing ( 222 ) said geometrical parameters with the geometrical parameters of the optical device having acquired the reference image sequence, and from the result of the comparison, determining ( 223, 224 ) whether the verification image sequence was acquired from at least one pre-recorded image of the individual.
1. A method for detecting fraud during biometric authentication, on the basis of a preregistered reference image sequence comprising at least one reference image of an individual acquired by an optical device during a preliminary enrollment step, the method comprising the steps of:
acquiring, with the same optical device, a sequence of candidate images comprising at least two candidate images of the individual,
estimating, from the sequence of candidate images, geometrical parameters of the optical device having acquired said sequence of candidate images on the individual, including a parameter equal to the ratio between the focal distance of the optical device having acquired said sequence and the size of the pixels of the sensor of said optical device,
comparing said geometrical parameters with geometrical parameters of the optical device having acquired the reference image sequence, including comparing the previously estimated parameter with the parameter equal to the ratio between the focal distance of the optical device having acquired the reference image sequence and the size of the pixels of the sensor of said optical device, and
from the result of the comparison, determining whether the sequence of candidate images was acquired from at least one pre-recorded image of the individual,
wherein each image sequence is a video comprising a plurality of images, and the images of each sequence are acquired for different relative positions of the individual with respect to the optical device.
2. The fraud detection method according to claim 1 , wherein, the estimated geometrical parameters are the same as those of the optical device having acquired the reference image sequence in the absence of fraud, and correspond to the geometrical parameters of an equivalent optical device having acquired the sequence of candidate images directly on the individual in the case of fraud by re-projection of a pre-recorded image, and wherein, if the estimated geometrical parameters are different from the geometrical parameters of the optical device having acquired the reference image sequence, the candidate image sequence is considered as having been acquired from a pre-recorded image, and the method comprises a fraud detection step.
3. The fraud detection method according to claim 1 , wherein the optical device comprises a sensor and optics, and the geometrical parameters of an optical device comprise the focal distance of the optics, the size of the pixels of the sensor, and the position of the projection of the centre of the optics on the sensor.
4. The fraud detection method according to claim 1 , wherein each optical device used for acquiring the sequence of candidate images and the reference image sequence is modeled as a pinhole camera such that the projection of a point P on an optical sensor of the optical device is provided by the relationship:
p=K [ Rt ] P
wherein K is the matrix of the intrinsic geometrical properties of the optical device, R is a matrix of rotation and t is a translation vector,
and the estimation of the geometrical parameters of the optical device having acquired the sequence of candidate images comprises the estimation of the parameters of the matrix K eq of an equivalent optical device having acquired said images directly on the individual, and the comparison with the parameters of the matrix K 0 of the optical device having acquired the candidate image sequence.
5. The fraud detection method according to claim 1 , further comprising a preliminary enrollment step wherein an individual is enrolled by acquiring, by means of the optical device, a reference image sequence of the individual comprising at least one image of the individual, and the optical device used for enrollment being the same as the optical device used for acquisition of the sequence of candidate images during biometric authentication.
6. A fraud detection system comprising an optical device for acquiring images and a processing unit comprising a processor and a memory, wherein the processor is configured to execute the following steps:
sending an instruction to the optical device to acquire a sequence of candidate images comprising at least two candidate images,
estimating, from the sequence of candidate images, geometrical parameters of the optical device having acquired said sequence of candidate images, including a parameter equal to the ratio between the focal distance of the optical device having acquired said sequence and the size of the pixels of the sensor of said optical device,
comparing said geometrical parameters with geometrical parameters of an optical device having acquired a preregistered reference image sequence, including comparing the previously estimated parameter with the parameter equal to the ratio between the focal distance of the optical device having acquired the reference image sequence and the size of the pixels of the sensor of said optical device, and
from the result of the comparison, determining whether the sequence of candidate images was acquired from at least one pre-recorded image,
wherein each image sequence is a video comprising a plurality of images, and the images of each sequence are acquired for different relative positions of an individual with respect to the optical device.
7. The fraud detection system according to claim 6 , said system being a personal computer, a mobile telephone or a digital tablet.
8. A non-transitory computer-readable storage medium comprising code instructions, when applied by a processor, for executing a method comprising steps of:
acquiring, with an optical device, a sequence of candidate images comprising at least two candidate images,
estimating, from the sequence of candidate images, geometrical parameters of the optical device having acquired said sequence of images, including a parameter equal to the ratio between the focal distance of the optical device having acquired said sequence and the size of the pixels of the sensor of said optical device,
comparing said geometrical parameters with reference geometrical parameters, including comparing the previously estimated parameter with the parameter equal to the ratio between the focal distance of the optical device having acquired the reference image sequence and the size of the pixels of the sensor of said optical device, and
from the result of the comparison, determining whether the sequence of candidate images was acquired from at least one pre-recorded image, and determining whether an individual having used the sequence of candidate images for a biometric authentication has attempted fraud,
wherein each image sequence is a video comprising a plurality of images, and the images of each sequence are acquired for different relative positions of the individual with respect to the optical device.