# US 12,450,942 B2 · System and method for performing face recognition > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,450,942 B2 - Document ID: US12450942B2 - Status: Granted - Kind: B2 - Application number: 17/785,445 - Filed: 2020-12-14 - Published: 2023-01-26 - Granted: 2025-10-21 - Earliest claimed: 2019-12-16 - Assignees: De-Identification Ltd. - Applicants: De-Identification Ltd. - Inventors: Eliran Kuta; Sella Blondheim; Gil Perry; Yoav Hacohen; Matan Ben Yosef; Or Gorodissky - Web page: https://patentgrep.com/patent/US12450942B2 ## CPC - G06F 21/32: using biometric data, e.g. fingerprints, iris scans or voiceprints - G06V 40/172: {Classification, e.g. identification} - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/08: Learning methods - G06N 3/09: Supervised learning - G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods} - G06V 10/7747: {Organisation of the process, e.g. bagging or boosting} - G06V 10/776: Validation; Performance evaluation - G06V 40/165: {using facial parts and geometric relationships} - G06V 40/171: {Local features and components; Facial parts (eye characteristics G06V40/18); Occluding parts, e.g. glasses; Geometrical relationships} - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A system and a method of performing face recognition may include: receiving a first facial image, depicting a first face, and a second facial image depicting a second face; applying an ML model on the first image, to produce a first representation vector, and applying the ML model on the second image to produce a second representation vector; comparing the first representation vector and the second representation vector; and associating the first face with the second face based on the comparison, where the ML model is trained to produce the representation vectors from the facial images, based on regions in the facial images that correspond to distinctiveness scores that are beneath a distinctiveness threshold. ## Claims ### Claim 1 A system for performing face recognition, the system comprising: a non-transitory memory device storing instruction code; and a processor associated with the memory device, and configured to execute the instruction code, whereupon execution of modules of said instruction code, the processor is configured to: receive a distinctiveness map, associating a distinctiveness score to one or more regions in an image depicting a first face, the distinctiveness map including a lowest distinctiveness score and one or more higher distinctiveness scores; select one or more regions of the first face, that are associated with a distinctiveness score higher than the lowest distinctiveness score, based on the distinctiveness map; modify the image at the one or more selected regions, to produce one or more modified facial images; and utilize a machine learning (ML) model to recognize a second face, wherein the ML model is trained to recognize the second face based on the one or more modified facial images. ### Claim 2 (depends on 1) The system of claim 1, wherein said distinctiveness score corresponds to prominence of an associated region in a process of identification of the first face by a face recognition algorithm. ### Claim 3 (depends on 1) The system of claim 1, wherein the processor is configured to receive the distinctiveness map by: a. identifying one or more regions in the first face; b. using a facial recognition system to determine a distinctiveness score of each identified region in relation to the first face; and c. assigning a score to each identified region, based on the distinctiveness score, to produce the distinctiveness map. ### Claim 4 (depends on 3) The system of claim 3, wherein the processor is further configured to: repeat steps (a) through (c) with a plurality of facial images, to obtain a respective plurality of distinctiveness maps; align the plurality of distinctiveness maps according to a reference geometry; and average the scores of corresponding regions of the plurality of distinctiveness maps, to produce an averaged distinctiveness map, and wherein the processor is configured to select one or more regions of the first face based on the averaged distinctiveness map. ### Claim 5 (depends on 1) The system of claim 1, wherein the processor is configured to modify the image at the one or more selected regions by omitting one or more regions of the first face that correspond to the highest scored regions in the distinctiveness map. ### Claim 6 (depends on 1) The system of claim 1, wherein the processor is configured to modify the image at the one or more selected regions by: receiving a second image depicting a second face; and replacing one or more regions of the first face, that correspond to the highest scored regions in the distinctiveness map, with corresponding regions of the second face. ### Claim 7 (depends on 1) The system of claim 1, wherein the processor is configured to modify the image at the one or more selected regions by applying one or more graphical modifications to one or more regions of the first face that correspond to the highest scored regions in the distinctiveness map. ### Claim 8 (depends on 1) The system of claim 1, wherein the processor is further configured to, during an inference stage: receive a source image depicting a source face and one or more target images depicting one or more respective target faces; infer the trained ML model on the source image to produce a source representation vector; infer the trained ML model on the one or more target images, to produce one or more respective target representation vectors; compare between the source representation vector and at least one target representation vector; and based on the comparison, determine the probability that the source representation vector and the at least one target representation vector correspond to the same face. ### Claim 9 A method of training an ML model to recognize a depicted face, the method comprising: receiving a first image depicting a first face; obtaining a distinctiveness map, associating a distinctiveness score to one or more regions in the first face, the distinctiveness map including a lowest distinctiveness score and one or more higher distinctiveness scores; selecting one or more regions of the first face, that are associated with a distinctiveness score that is higher than the lowest distinctiveness score, based on the distinctiveness map; performing one or more modifications of the selected one or more regions of the first image, to produce one or more modified images; and training the ML model to recognize the depicted face, using the one or more modified images as a training dataset, and using the first image as supervisory data. ### Claim 10 (depends on 9) The method of claim 9, wherein obtaining the distinctiveness map comprises: a. identifying one or more regions in the first face; b. using a facial recognition system to determine a distinctiveness score of each identified region in relation to the first face; and c. assigning a score to each identified region, based on the distinctiveness score, to produce the distinctiveness map. ### Claim 11 (depends on 10) The method of claim 10, further comprising repeating steps (a) through (c) with a plurality of facial images, to obtain a respective plurality of distinctiveness maps; aligning the plurality of distinctiveness maps according to a reference geometry; and averaging the scores of corresponding regions of the plurality of distinctiveness maps, to produce an averaged distinctiveness map, and wherein selecting one or more regions of the first face is based on the averaged distinctiveness map. ### Claim 12 (depends on 9) The method of claim 9, wherein performing the one or more modifications of the selected one or more regions comprises omitting one or more regions of the first face that correspond to the highest scored regions in the distinctiveness map. ### Claim 13 (depends on 9) The method of claim 9, wherein performing the one or more modifications of the selected one or more regions comprises: receiving a second image depicting a second face; and replacing one or more regions of the first face, that correspond to the highest scored regions in the distinctiveness map, with corresponding regions of the second face. ### Claim 14 (depends on 9) The method of claim 9, wherein performing the one or more modifications of the selected one or more regions comprises applying one or more graphical modifications to one or more regions of the first face that correspond to the highest scored regions in the distinctiveness map. ### Claim 15 (depends on 9) The method of claim 9, further comprising, during an inference stage: receiving a source image depicting a source face and one or more target images depicting one or more respective target faces; inferring the trained ML model on the source image to produce a source representation vector; inferring the trained ML model on the one or more target images, to produce one or more respective target representation vectors; comparing between the source representation vector and at least one target representation vector; and based on the comparison, determining the probability that the source representation vector and the at least one target representation vector correspond to the same face. ### Claim 16 A method of performing face recognition by at least one processor, the method comprising: receiving a first facial image, depicting a first face, and a second facial image depicting a second face; applying an ML model on the first image, to produce a first representation vector, and applying the ML model on the second image to produce a second representation vector; comparing the first representation vector and the second representation vector; and associating the first face with the second face based on the comparison, wherein the ML model is trained to produce the representation vectors from the facial images, based on regions in the facial images that correspond to distinctiveness scores that are beneath a distinctiveness threshold. ### Claim 17 (depends on 16) The method of claim 16, wherein training the ML model comprises: receiving a first image depicting a face; obtaining a distinctiveness map, associating a distinctiveness score to one or more regions in the depicted face; selecting one or more regions of the face, that are associated with a distinctiveness score higher than the lowest distinctiveness score, based on the distinctiveness map; performing at least one modification of the first image, to produce one or more second images comprising modified versions of the selected regions; providing the one or more second images as a training dataset to the ML model; and training the ML model to recognize the depicted face, using the first image as supervisory input, wherein said distinctiveness score corresponds to prominence of an associated region, in a process of identification of the face by a face recognition algorithm.