SYSTEM FOR SECURE FACE IDENTIFICATION (SCIFI) AND METHODS USEFUL IN CONJUNCTION THEREWITH

• US 20110026781A1
• Filed: 07/01/2010
• Published: 02/03/2011
• Est. Priority Date: 07/02/2009
• Status: Active Grant

• Current Assignee: Carmel-Haifa University Economic corp. Ltd. (University of Haifa)
• Original Assignee: Carmel-Haifa University Economic corp. Ltd. (University of Haifa)
• Inventors: Pinkas, Binyamin, Osadchy, Margarita, Jarrous, Ayman, Moskovich, Boaz
• Granted Patent: US 8,542,886 B2
• Time in Patent Office: Days
• US Class Current: 382/118
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions
• CPC Class Codes: H04L 9/008 involving homomorphic encryption

Abstract

A method for privacy-retaining face identification comprising dividing at least one input face into a first set of patches, providing an assortment of patch profiles for each patch in the first set of patches, each patch profile in the assortment being associated with a unique index thereby to define an assortment of unique indices, for each individual patch in the input face, finding at least one patch profile within the assortment of patch profiles which most resembles the individual patch, thereby to define a first set of indices from among the assortment of unique indices whose corresponding patch profiles most resemble the input face'"'"'s patches respectively, and using a privacy preserving computation to generate a physical output indicative of a comparison of the input face and at least one target face by comparing first and second functions of the first set of indices and of a second set of indices respectively, the second set of indices corresponding to patch profiles which most resemble a second set of patches into which the target face has been divided.

Claims

1. A method for privacy-retaining face identification comprising:

dividing at least one input face into a first set of patches;

providing an assortment of patch profiles for each patch in said first set of patches, each patch profile in said assortment being associated with a unique index thereby to define an assortment of unique indices;

for each individual patch in the input face, finding at least one patch profile within said assortment of patch profiles which most resembles said individual patch, thereby to define a first set of indices from among said assortment of unique indices whose corresponding patch profiles most resemble said input face'"'"'s patches respectively; and

using a privacy preserving computation to generate a physical output indicative of a comparison of said input face and at least one target face by comparing first and second functions of said first set of indices and of a second set of indices respectively, said second set of indices corresponding to patch profiles which most resemble a second set of patches into which said target face has been divided.

4. A method according to claim 1 wherein said respective functions comprise binary strings generated from said first and second sets of indices respectively and wherein a Hamming distance is used to compare said binary strings.

5. A method according to claim 1 wherein said assortment of patch profiles comprises a finite assortment.

9. A method according to claim 1 and also comprising computing a first binary string representing said first set of indices which includes a “

1”

at each position within said binary string which corresponds to an index in said first set of indices and a “

0”

at each position within said binary string which does not correspond to any index in said first set of indices.

10. A method according to claim 1 also comprising computing a second binary string representing said second set of indices which includes a “

1”

at each position within said second binary string which corresponds to an index in said second set of indices and a “

0”

at each position within said second binary string which does not correspond to any index in said second set of indices.

11. A method according to claim 9 wherein said function of said first set of indices comprise a homomorphic encryption of said first binary string.

12. A method according to claim 10 wherein said function of said second set of indices comprise a homomorphic encryption of said second binary string.

13. A method according to claim 1 wherein said using of said privacy preserving computation comprises determining similarity between said first and second functions of said first and second sets of indices respectively at least partly based on a determination of whether the number of individual bit differences in at least respective portions of respective binary strings representing said functions is less than a certain threshold.

14. A method according to claim 1 wherein said finding of at least one patch profile does not divulge any information other than whether or not the input face matches said target face.

15. A method according to claim 1 which determines whether said at least one input face matches said at least one target face without divulging the input face to the server and without divulging the target face to the client.

17. A method according to claim 1 also comprising computing at least one binary string representing at least one set of indices from among said first and second sets of indices, whose binary string includes a “

1”

at each position within said binary string which corresponds to an index in said at least one set of indices and a “

0”

at each position within said binary string which does not correspond to any index in said at least one set of indices.

18. A method according to claim 1 wherein said physical output comprises a command affecting access to a restricted area.

19. A method according to claim 1 wherein said physical output comprises an output on a computerized display.

20. A method according to claim 1 and also comprising imaging a human'"'"'s face to obtain a computerized representation of at least one face.

21. A method according to claim 1 wherein said first and second functions both comprise the unity function such that said privacy preserving computation compares said first set of indices to said second set of indices.

2. A method according to claim I and also comprising performing the following:

dividing at least one target face into a second set of patches; and

for each individual patch in the target face, finding at least one patch profile within said assortment of patch profiles which most resembles said individual patch, thereby to define the second set of indices from among said assortment of unique indices whose corresponding patch profiles most resemble said target face'"'"'s patches respectively.

3. A method according to claim 2 wherein a server processor having a library of at least one target face to be identified divides said target face into said second set of patches and defines said second set of indices and a client processor divides said input face into said second set of patches and defines said second set of indices.

6. A method according to claim 3 wherein said assortment of patch profiles for each patch in said first and second sets of patches comprises a public database available to said server processor and to said client processor.

7. A method according to claim 3 wherein said using includes performing said privacy preserving computation without conveying any information regarding said input face, other than a function of said first set of indices, from said client processor to said server processor.

8. A method according to claim 3 wherein said using includes performing said privacy preserving computation without conveying any information regarding said input face, other than a function of said second set of indices, from said server processor to said client processor.

16. A method according to claim 3 wherein only one of said server processor and said client processor learns whether or not said at least one input face matches said at least one target face.

22. A system for privacy-retaining face identification operative in association with a server side having a library of at least one target face to be identified and a client side having at least one input face whose resemblance to said at least one target face is to be determined, the system including:

at least one face patcher dividing the input face into a first set of patches and dividing the target face into a second set of patches;

a database including an assortment of patch profiles for each patch in said first and second sets of patches, each patch profile in said assortment being associated with a unique index thereby to define an assortment of unique indices;

a discrete face analyzer finding, for each individual patch in the input face and in the target face, at least one patch profile within said assortment of patch profiles which most resembles said individual patch, thereby to define a first set of indices from among said assortment of unique indices whose corresponding patch profiles most resemble said input face'"'"'s patches respectively and a second set of indices whose corresponding patch profiles most resemble said target face'"'"'s patches respectively; and

a privacy preserving computation machine operative to compare first and second functions of said first and second sets of indices respectively.

23. A computer program product, comprising a computer usable medium having a computer readable program code embodied therein, said computer readable program code adapted to be executed to implement a method for privacy-retaining face identification comprising:

dividing at least one input face into a first set of patches;

providing an assortment of patch profiles for each patch in said first set of patches, each patch profile in said assortment being associated with a unique index thereby to define an assortment of unique indices;

for each individual patch in the input face, finding at least one patch profile within said assortment of patch profiles which most resembles said individual patch, thereby to define a first set of indices from among said assortment of unique indices whose corresponding patch profiles most resemble said input face'"'"'s patches respectively; and

using a privacy preserving computation to generate a physical output indicative of a comparison of said input face and at least one target face by comparing first and second functions of said first set of indices and of a second set of indices respectively, said second set of indices corresponding to patch profiles which most resemble a second set of patches into which said target face has been divided.