Granted Patent B2

US 12,346,422 · App. 18/080,590 · Granted Jul 1, 2025

Inventor: Arun Vemury (North Bethesda, MD)

Assignee: The Government of the United States of America, as represented by the Secretary of Homeland Security

G06F21/32G06F21/602G06F21/6245G06V40/168G06V40/172G06V40/50H04L9/008H04L9/3231

Abstract

Claims (91)

1. A method comprising:

a system controller comprising:

generating a homomorphic encryption (HE) encrypted reference biometric FV table for a third-party computer resource;

distributing the HE encrypted reference biometric feature vector (FV) table to a third-party computer resource;

generating a controller-generated fail-over exception table for a biometric capture and a HE encrypted distribution unit;

distributing the controller-generated fail-over exception table to the biometric capture and the HE encrypted distribution unit;

the system controller comprising a processor, instruction memory coupled to one another through a bus;

generating a reference biometric HE vector table containing biometric information;

generating a biometric reference gallery of integer N reference images, biographic reference images RG(i), i=1, N;

include initializing an “i” index to integer 1; and

computing a feature vector FV(RG(i)) from the RG(i), at i equal to integer 1.

2. The method of claim 1 wherein computing the feature vector includes applying an orthogonal basis transformation selected from a list consisting essentially of discrete cosine transform (DCT), discrete wavelet transform (DWT), Walsh Transform, and Walsh Hadamard Transform (WHT).

3. The method of claim 1 comprising computing the feature vector, FV(RG(i)), of RG(i) to HE encrypt the feature vector FV(RG(i)) to obtain HE{FV(RG(i))}.

4. The method of claim 3 comprising collision checking a controller maintained HE{FV(RG)} table to determine whether HE{FV(RG(i))} matches an already present HE encrypted FV in the controller maintained (CNTLR) HE{FV(RG)} table; wherein the CNTLR means “controller maintained”.

5. The method of claim 3 comprising:

providing an assumed starting state including an empty CNTLR HE{FV(RG)} table; wherein collision checking produces a negative result, no collision, because the assumed starting state includes the empty CNTRL HE{FV(RG)} table;

executing a computer software loop comprising the steps of:

inserting the HE{FV(RG(1))} in the CNTLR HE{FV(RG)} table;

computing feature vector for RG(2);

HE encrypting FV(RG(2));

collision checking HE{FV(RG(2))} the CNTLR HE{FV(RG)} table;

inserting HE{FV(RG(2))} in the CNTLR HE{FV(RG)} table; and

incrementing the i index by integer 1;

wherein the CNTLR means “controller maintained”.

6. The method of claim 5 comprising:

repeating the loop until “I” is at an arbitrary (BX) “BX;”

identifying a HE encryption collision, wherein the CNTLR HE{FV(RG)} table already includes a HE encrypted feature vector that matches HE{FV(RG(BX))}; wherein a HE vector and a current HE{FV(RG(BX))} are identical, and the HE vector is the HE encryption of a FV of an earlier processed reference gallery image RG(X);

removing an already present HE{FV(RG(BX))} from the CNTLR HE{FV(RG)} table; and

inserting or adding HE{FV(RG(BX))} to a controller maintained exception table (Exception Table EXC_FV(RG)).

7. The loop of claim 6 comprising:

adding HE{FV(RG(BX))} to the controller maintained exception table (Exception Table EXC_FV(RG)); and

determining if there are any more RGs in the biometric reference gallery.

8. The loop of claim 7 wherein the CNTRL Exception Table EXC_FV(RG) is the controller stored construction of a distributed exception table configured to implement a Central HE Vector Exception Table.

9. The loop of claim 7 comprising removing an already present HE{FV(RG(BX))} from the CNTLR HE{FV(RG)} table; wherein removing means flagging the already present HE{FV(RG(BX)) as a collision value, for use by operations in subsequent loops.

10. The loop of claim 9 comprising:

repeating operations in the flow as a sequence of loops;

incrementing index i by integer 1; and either;

adding another HE{FV(RG(i))} to the CNTLR HE{FV(RG)} table; or

removing another HE{FV(RG(i))} from the CNTLR HE{FV(RG)} table and adding the HE{FV(RG(i))} to the CNTRL Exception Table EXC_FV(RG) until an operations branch determines there are no more reference galleries (RGs) in the biometric reference gallery.

11. The loop of claim 10 comprising:

communicating a constructed CNTLR HE{FV(RG)} table to the third-party computer resource; and

communicating the constructed CNTRL Exception Table EXC_FV(RG) to a-biometric capture, exception protective (EP) HE encryption and distribution (ECDS) unit.

12. A method for secure privacy-protective HE distribution and persistent encryption HE secured, non-decrypting distributed biometric processing, with a priori distribution of exceptions, for local cleartext failover comprising:

providing a third-party computer resource; a biometric capture, HE encrypted distribution station; a biometric processing exception detection and failover logic comprising a central homomorphic encrypted (HE) vector exception table; a HE{FV} validity logic; a FO VF-ID logic, a HEFV table; a reference HE FV table, for HE{FV(RG(j))}, j=1, a FV(RG(i)) logic; a HE{FV(RG(i))} logic; and a HE vector table collision detect, update logic; the third-party resource comprising a processor, instruction memory coupled to one another through a bus; FV is feature vector; FO is failover; VF-ID is failover verification and identification; RG is biographic reference image;

a HE domain classifier configuring logic generating a cleartext biometric reference gallery; the cleartext biometric reference gallery comprising HE encryption of distinguishing information regarding M biometric reference images; M is a natural number;

setting an index i to 1;

retrieving the RG(1) biometric reference image from the cleartext biometric reference gallery;

feeding RG(1) to obtain a HE encrypted FV for RG(1);

a HE vector collision detect logic receiving HE{FV(RG(1))} and checking to determine if a reference HE FV table already includes a HE vector of identical value;

loading HE{FV(RG(1))} into the reference HE FV table; and

constructing the reference HE FV table for performing another iteration.

13. The method of claim 12 comprising the steps of selecting another biometric reference image from the cleartext biometric reference gallery;

incrementing the index i from 1 to 2;

retrieving RG(2) biometric reference image from the cleartext biometric reference gallery;

feeding RG(2) to HE FV logic to obtain HE{FV(RG(2))};

checking the reference HE FV table to determine whether the reference HE FV table already includes a HE encrypted FV having a value HE{FV(RG(2))}; and

loading HE{FV(RG(2))} into the reference HE FV table if the determination is negative.

14. A method of distributed privacy protective, HE persistent encryption locked, non-decrypting biometric identification and verification processing, with exception failover backup for exception failover augmented, homomorphic encrypted (HE) distributing, end-to-endpoint persistent encryption, and distributed HE domain non-decrypting, privacy-protective biometric processing of a current user K comprising:

receiving an HE encrypted reference gallery of N HE encrypted reference files from the current user K; each of the N HE encrypted reference files corresponding to a verified reference identity; N is a natural number; K is a natural number;

receiving HE encrypted feature vectors representing captured facial image information from a biometric capture EP HE ECDS unit in response to user interactions with the biometric capture EP HE ECDS unit; EP is exception protective; ECDS is encryption and distribution, performing HE domain, non-decrypting N+1 class classification of the HE encrypted feature vectors, using a vector similarity algorithm;

detecting an exception; and

providing failover processing by the biometric capture, EP HE ECDS unit in response to detecting the exception.

15. The method of claim 14 comprising:

a-priori initializing of a “k” index at integer 1, and providing an initial empty state of an accumulated HE{FV} memory; FV is feature vector and receiving biographic information (BGH(k)) from an application on a smart phone operated by the user K with the biometric capture EP HE ECDS unit.

16. The method of claim 15 wherein the biographic information BGH(k) includes a full name, date of birth, and mailing address.

17. The method of claim 15 comprising:

assigning a temporary identifier (TID(K)); and

storing a correspondence, represented as TID(k)↔BGH(k), in a memory of a kiosk; wherein k corresponds to a user k.

18. The method of claim 15 wherein the biometric capture EP HE ECDS unit is a camera configured to capture a row-by-column pixel array in red green blue (RGB) of the user's face.

19. The method of claim 18 comprising:

capturing biometric information (BM(k));

storing a copy of the captured biometric information BM(k) for potential use in a fail-over process;

computing a feature vector, FV(BGH(k)) for BM(k);

storing a copy of the FV(BGH(k)); wherein FV(BGH(k)) includes representation of biometric features unique to k with a substantially smaller file size than the original pixel array;

HE encrypting the feature vector FV(BM(k)) to obtain HE{FV(BM(k))};

comparing HE{FV(BM(k))} to the accumulated HE{FV(BM)} memory of previously used HE{FV(BM(k))}; wherein a result of the compare is a positive or negative outcome;

inserting the HE{FV(BM(k))} into the accumulated HE{FV(BM)} memory; and distributing HE{FV(BM(k))} to third-party computer resources for HE domain failover verification and identification (VF-ID) processing if the comparing is a negative outcome; and

referring to the user for which the accumulated HE{FV(BM))} value which is identical to the current HE{FV(BM(k))} was generated as a “c th ” user if the comparing is a positive outcome.

20. The method of claim 19 comprising:

determining whether biographical information for the c th user matches the biographical information for a current k th user;

if the determination is positive, verifying the current k th user as the c th user; and

if the determination is positive, uploading HE{FV(BM(k))} to the third-party computer resources; and

if the determination is negative, flagging HE{FV(BM(k))} as invalid, an exception based on a collision; and perform a system-internal cleartext biometric process using a cleartext of the captured biometric information BM(k) or a cleartext of the feature vector FV(BM(k)).

21. The method of claim 20 comprising:

inserting the HE{FV(BM(k))} into the accumulated HE{FV(BM)} memory; and distribute HE{FV(BM(k))} to third-party computer resources for HE domain failover verification and identification (VF-ID) processing if the comparing is a negative outcome; and

if HE{FV(BM(k))} is distributed to the third-party computer resource:

receiving the third-party computer resource HE processing result, which includes TID(k); and

determining whether a match was found;

if a match was found, ending the method with respect to a most recent user; and

if a match was not found, performing system internal cleartext biometric processing, using a cleartext form of BM(k) or a cleartext form of FV(BM(k)).

22. The method of claim 1 , wherein the generated biometric reference gallery contains facial images, fingerprint images, or iris images.

Assignments (1)

Continuity (2)

References Cited (64)