Privacy Preserving Biometric Verifiability

Why is SenseCrypt non-biometric in nature?

Traditional Face Recognition

As we have seen in the previous section, traditional Face Recognition depends on generating and storing a biometric template during enrollment and then, subsequently, generating and comparing a new template during verification.

Since two generated feature vector / templates can be compared, then if one is stored in a database operated by company A, and the other is stored in a database operated by company B, a comparison can be made between them to determine if it is the same person.

This is despite the fact that company A and company B may be unrelated to each other. This violates the principle of Unlinkability in a Privacy by Design framework.

Privacy Preservation

The ability to compare feature vectors / templates with one another in traditional biometrics is what makes the stored data biometric in nature.

To enable privacy preservation, a system should be able to generate any number of different data structures (akin to feature vectors / templates) from a single image. If it does so, then the data structures thus generated cannot be compared to one another.

Since there is no way to compare the data structures generated by a privacy preserving framework, if they were stored in separate databases, there would be no way to find out that the data structures correspond to the same person. This satisfies the principle of Unlinkability in a Privacy by Design framework. This is also what makes the data structures non-biometric in nature.

Biometric Verifiability

We have seen how two Privacy Preserving data structures generated from exactly the same data cannot be compared to each other to determine any kind of similarity.

But, given a Biometric Sample (such as a facial image), can it be determined that the Privacy Preserving data structure was generated from a similar biometric sample?

As it turns out, the above is made possible through the SenseCrypt algorithm.

Thus, even though the data structures (SensePrints) generated by SenseCrypt algorithm are non biometric in nature, they can be used for Biometric Verification.