System and method for robust palm liveness detection using variations of images

1. A method implemented on at least one processor, a memory, and a communication platform enabling connection to a network for detecting liveness, comprising:

receiving an input image with visual information claimed to represent a palm of a person;

accessing one or more models, each of which specifies conditions to generate a variation of the input image;

generating one or more variations of the input image based on the one or more models;

with respect to each image of the one or more variations and the input image,

determining a region of interest (ROI) of the image with biometric information enclosed therein, wherein the ROI satisfies a predetermined dimension requirement, and

generating a liveness decision indicative of whether the ROI represents live biometric information; and

detecting liveness of the visual information in the input image based on the liveness decisions made with respect to the input image and the one or more variations of the input image.

2. The method of claim 1 , wherein the one or more variations include at least one of:

a first variation of the input image corresponding to a focused image of a region of the input image;

a second variation of the input image generated based on the input image with a different lighting condition; and

a third variation of the input image generated based on the input image with modified intensities.

3. The method of claim 2 , wherein the biometric information comprises a palm.

4. The method of claim 3 , wherein the step of determining the ROI comprises:

processing the image to identifying an initial ROI that encloses the palm;

assessing the initial ROI with respect to the predetermined dimension requirement;

if the initial ROI does not satisfy the predetermined dimension requirement,

determine an adjustment needed to make the initial ROI meet the predetermined dimension requirement, and

generating the ROI by applying the adjustment to the initial ROI; and

if the initial ROI satisfies the predetermined dimension requirement, using the initial ROI as the ROI.

5. The method of claim 3 , wherein the step of generating the liveness decision with respect to the ROI comprises:

invoke a plurality of individual liveness detectors, each of which is provided to detect one type of fake palm based on visual information in the ROI in accordance with a detection model obtained via machine learning for the type of fake palm, wherein a presence of a fake palm indicates that the visual information in the ROI does not represent a live palm;

receiving a liveness detection result from each of the plurality of individual liveness detectors;

combining the liveness detection results from the plurality individual liveness detectors to derive the liveness decision with respect to the ROI.

6. The method of claim 5 , wherein the step of combining the liveness detection results from the plurality of individual liveness detectors is performed based on an integration model, obtained via machine learning.

7. The method of claim 3 , wherein the step of determining the liveness of the visual information in the input image comprises:

accessing liveness decision integration models for integrating the liveness decisions made with respect to the input image and the one or more variations;

analyzing corresponding confidence scores associated with the liveness decisions with respect to the input image and the one or more variations;

integrating the liveness decisions based on the confidence scores associated with the liveness decisions and in accordance with the liveness decision integration models;

generating an output indicating the liveness of the visual information in the input image based on a result of the integrating.

8. Machine readable and non-transitory medium having information recorded thereon for detecting liveness, wherein the information, when read by the machine, causes the machine to perform the following steps:

receiving an input image with visual information claimed to represent a palm of a person;

accessing one or more models, each of which specifies conditions to generate a variation of the input image;

generating one or more variations of the input image based on the one or more models;

with respect to each image of the one or more variations and the input image,

determining a region of interest (ROI) of the image with biometric information enclosed therein, wherein the ROI satisfies a predetermined dimension requirement, and

generating a liveness decision indicative of whether the ROI represents live biometric information; and

detecting liveness of the visual information in the input image based on the liveness decisions made with respect to the input image and the one or more variations of the input image.

9. The medium of claim 8 , wherein the one or more variations include at least one of:

a first variation of the input image corresponding to a focused image of a region of the input image;

a second variation of the input image generated based on the input image with a different lighting condition; and

a third variation of the input image generated based on the input image with modified intensities.

10. The medium of claim 9 , wherein the biometric information comprises a palm.

11. The medium of claim 10 , wherein the step of determining the ROI comprises:

processing the image to identifying an initial ROI that encloses the palm;

assessing the initial ROI with respect to the predetermined dimension requirement;

if the initial ROI does not satisfy the predetermined dimension requirement,

determine an adjustment needed to make the initial ROI meet the predetermined dimension requirement, and

generating the ROI by applying the adjustment to the initial ROI; and

if the initial ROI satisfies the predetermined dimension requirement, using the initial ROI as the ROI.

12. The medium of claim 10 , wherein the step of generating the liveness decision with respect to the ROI comprises:

invoke a plurality of individual liveness detectors, each of which is provided to detect one type of fake palm based on visual information in the ROI in accordance with a detection model obtained via machine learning for the type of fake palm, wherein a presence of a fake palm indicates that the visual information in the ROI does not represent a live palm;

receiving a liveness detection result from each of the plurality of individual liveness detectors;

combining the liveness detection results from the plurality individual liveness detectors to derive the liveness decision with respect to the ROI.

13. The medium of claim 12 , wherein the step of combining the liveness detection results from the plurality of individual liveness detectors is performed based on an integration model, obtained via machine learning.

14. The medium of claim 10 , wherein the step of determining the liveness of the visual information in the input image comprises:

accessing liveness decision integration models for integrating the liveness decisions made with respect to the input image and the one or more variations;

analyzing corresponding confidence scores associated with the liveness decisions with respect to the input image and the one or more variations;

integrating the liveness decisions based on the confidence scores associated with the liveness decisions and in accordance with the liveness decision integration models;

generating an output indicating the liveness of the visual information in the input image based on a result of the integrating.

15. A system for detecting liveness, comprising:

a palm image variation generator implemented by a processor and configured for:

receiving an input image with visual information claimed to represent a palm of a person,

accessing one or more models, each of which specifies conditions to generate a variation of the input image, and

generating one or more variations of the input image based on the one or more models; and

a mechanism implemented by the processor and configured for

with respect to each image of the one or more variations and the input image,

determining a region of interest (ROI) of the image with biometric information enclosed therein, wherein the ROI satisfies a predetermined dimension requirement, and

generating a liveness decision indicative of whether the ROI represents live biometric information; and

detecting liveness of the visual information in the input image based on the liveness decisions made with respect to the input image and the one or more variations of the input image.

16. The system of claim 15 , wherein the one or more variations include at least one of:

a first variation of the input image corresponding to a focused image of a region of the input image;

a second variation of the input image generated based on the input image with a different lighting condition; and

a third variation of the input image generated based on the input image with modified intensities.

17. The system of claim 16 , wherein the biometric information comprises a palm.

18. The system of claim 17 , wherein the step of determining the ROI comprises:

processing the image to identifying an initial ROI that encloses the palm;

assessing the initial ROI with respect to the predetermined dimension requirement;

if the initial ROI does not satisfy the predetermined dimension requirement,

determine an adjustment needed to make the initial ROI meet the predetermined dimension requirement, and

generating the ROI by applying the adjustment to the initial ROI; and

if the initial ROI satisfies the predetermined dimension requirement, using the initial ROI as the ROI.

19. The system of claim 17 , wherein the palm liveness detector comprises:

a plurality of individual liveness detectors, each of which is implemented by the processor and configured to

detect one type of fake palm based on visual information in the ROI in accordance with a detection model obtained via machine learning for the type of fake palm, wherein a presence of a fake palm indicates that the visual information in the ROI does not represent a live palm, and

generate a liveness detection result based on whether the type of fake palm is detected; and

a fake palm detection integrator implemented by the processor and configured for combining the liveness detection results from the plurality individual liveness detectors to derive the liveness decision with respect to the ROI.

20. The system of claim 19 , wherein the liveness detection results from the plurality of individual liveness detectors are combined based on an integration model, obtained via machine learning.

21. The system of claim 17 , wherein the step of determining the liveness of the visual information in the input image comprises:

accessing liveness decision integration models for integrating the liveness decisions made with respect to the input image and the one or more variations;

analyzing corresponding confidence scores associated with the liveness decisions with respect to the input image and the one or more variations;

integrating the liveness decisions based on the confidence scores associated with the liveness decisions and in accordance with the liveness decision integration models;

generating an output indicating the liveness of the visual information in the input image based on a result of the integrating.