Systems and methods for privacy-enabled biometric processing
In one embodiment, a set of feature vectors can be derived from any biometric data, and then using a deep neural network (“DNN”) on those one-way homomorphic encryptions (i.e., each biometrics' feature vector) can determine matches or execute searches on encrypted data. Each biometrics' feature vector can then be stored and/or used in conjunction with respective classifications, for use in subsequent comparisons without fear of compromising the original biometric data. In various embodiments, the original biometric data is discarded responsive to generating the encrypted values. In another embodiment, the homomorphic encryption enables computations and comparisons on cypher text without decryption. This improves security over conventional approaches. Searching biometrics in the clear on any system, represents a significant security vulnerability. In various examples described herein, only the one-way encrypted biometric data is available on a given device. Various embodiments restrict execution to occur on encrypted biometrics for any matching or searching.
1. A privacy-enabled identification system comprising:
at least one processor operatively connected to a memory;
a classification component executed by the at least one processor, including at least one classification model having a neural network that is a first member of at least one linked pair of models, the neural network configured to:
classify a first set of distance measurable encrypted feature vectors generated by a second member of the at least one linked pair of models and respective label inputs during training;
define at least a plurality of identification classes, each class of the plurality of identification classes corresponding to a respective entity and assigned a respective label associated with the respective entity for subsequent identification;
return a label associated with a respective identification class for entity identification responsive to matching to the respective identification class and an unknown result responsive to failing to match to the plurality of identification classes during prediction, responsive to input of a newly generated distance measurable encrypted feature vector;
a feature vector generation component comprising at least a pre-trained feature model having a pre-trained convolutional neural network (“CNN”) as the second member of the at least one linked pair of models, the pre-trained CNN is trained to generate distance measurable encrypted feature vectors for any entity, wherein generation of distance measurable encrypted feature vectors occurs independent of any label assigned, as an output of at least one layer in the pre-trained CNN responsive to input of a plain text identification input to the pre-trained CNN, wherein the at least one linked pair of networks are linked based on a biometric data type.
2. The system of claim 1 , wherein the classification component is configured to predict a match to a label responsive to input of plaintext identification information of a single identification data type.
3. The system of claim 1 , wherein the classification component further comprises a second classification model having a neural network that is a first member of a second linked pair of networks, the second classification network is configured to accept distance measurable encrypted feature vectors from a second pre-trained neural network that is a second member of the second linked pair of models configured to generates a second set of distance measurable encrypted feature vectors on a second identification data type.
4. The system of claim 1 , wherein the pre-trained convolutional neural network includes an output generation layer which provides homomorphic distance measurable encrypted feature vectors.
5. The system of claim 1 , wherein the system is configured to instantiate multiple classification models and multiple neural networks, classification model associated with at least one different identification data type relative to another classification model, and classify new respective inputs of respective distance measurable encrypted feature vectors based on respective identification data type.
6. The system of claim 5 , wherein the system is configured to execute a voting procedure to increase accuracy of identification based on multiple identification data inputs or multiple types of biometric input.
7. The system of claim 5 , wherein the system is configured to maintain at least an executing copy of the neural network and an updatable copy of the neural network that can be locked or put in an offline state to enable training operations while the executing copy of the neural network executes any classification requests.
8. The system of claim 1 , wherein the classification component is configured with a plurality of modes of execution, including an enrollment mode configured to accept a label for identification and input feature vectors for an entity produced from the feature vector generation component.
9. The system of claim 1 , wherein the classification component is configured to predict a match to an existing label or to return an unknown result based on feature vectors enrolled in the neural network.
10. The system of claim 1 , wherein the feature generation component is pre-trained prior to enrollment of entities to be identified, and the classification component is trained subsequent to enrollment of the respective entity to be identified.
11. The system of claim 1 , wherein the pre-trained feature model and respective pre-trained neural networks are pre-trained prior to enrollment of any entity to be identified or authenticated.
12. A computer implemented method for privacy-enabled identification comprising:
instantiating, by at least one processor, at least one classification model having a neural network that is a first member of at least one linked pair of models;
classifying, by the neural network, a first set of distance measurable encrypted feature vectors generated by a second member of the at least one linked pair of models and respective label inputs during training;
defining, by the neural network, at least a plurality of identification classes, each class of the plurality of identification classes corresponding to a respective entity and assigned a respective label associated with the respective entity for subsequent identification;
returning, by the at least one processor, a label associated with a respective identification class for entity identification responsive to matching to the respective identification class and returning an unknown result responsive to failing to match to the plurality of identification classes during prediction, responsive to input of a newly generated distance measurable encrypted feature vector;
instantiating, by the at least one processor, a feature vector generation component comprising at least a pre-trained feature model having a pre-trained convolutional neural network (“CNN”) as the second member of the at least one linked pair of models, the pre-trained CNN trained to generate distance measurable encrypted feature vectors for any entity, wherein generation of distance measurable encrypted feature vectors occurs independent of any label assigned, as an output of at least one layer in the pre-trained CNN responsive to input of a plain text identification input to the pre-trained CNN, wherein the at least one linked pair of networks are linked based on a biometric data type.
13. The method of claim 12 , wherein the method further comprises predicting a match to a label responsive to input of plaintext identification information of a single identification data type.
14. The method of claim 12 , wherein the method further comprises instantiating a second classification model having a neural network that is a first member of a second linked pair of networks, the second classification network configured to accept distance measurable encrypted feature vectors from a second pre-trained neural network that is a second member of the second linked pair of models that generates a second set of distance measurable encrypted feature vectors on a second identification data type.
15. The method of claim 12 , wherein the method further comprises generating homomorphic distance measurable encrypted feature vectors from an output generation layer of the pre-trained convolutional neural network includes.
16. The method of claim 12 , wherein the method further comprises:
instantiating multiple classification models and multiple neural networks, and each classification model associated with at least one different identification data type relative to another classification model; and
classifying new respective inputs of respective distance measurable encrypted feature vectors based on respective identification data type.
17. The method of claim 16 , wherein the method further comprises executing a voting procedure to increase accuracy of identification based on multiple identification data inputs or multiple types of biometric input.
18. The method of claim 16 , wherein the method further comprises maintaining at least an executing copy of the neural network and an updatable copy of the neural network that can be locked or put in an offline state to enable training operations while the executing copy of the neural network executes any classification requests.
19. The method of claim 12 , wherein the classification component is configured with a plurality of modes of execution, and the method further comprises accepting a label for identification and input feature vectors for an entity produced from the feature vector generation component as part of an enrollment mode.
20. The method of claim 12 , wherein the method further comprises predicting a match to an existing label or returning an unknown result based on feature vectors enrolled in the neural network.
21. The method of claim 12 , wherein the method further comprises pre-training the feature generation component prior to enrollment of entities to be identified, and training the classification component subsequent to enrollment of the respective entity to be identified.
22. The method of claim 12 , wherein the method further comprises pre-training the pre-trained feature model and respective pre-trained neural networks prior to enrollment of any entity to be identified or authenticated.