Granted Patent B2

US 12,664,313 · App. 18/524,419 · Granted Jun 23, 2026

Inventors: Kieran Fraser (Dublin, IE); Liubov Nedoshivina (Dublin, IE); Anisa Halimi (Dublin, IE); Stefano Braghin (Dublin, IE)

Assignee: International Business Machines Corporation

G06F21/6263G16H40/67

Abstract

Claims (39)

1 . A method for visual content privatization, the method comprising:

receiving visual content associated with a subject;

altering the visual content using at least one or more image perturbations or one or more adversarial patches in response to a practitioner requesting an external consultation from a third party, wherein the visual content is altered within a privacy-utility system comprised of a feature recognition component and a visual content altering component, wherein the visual content altering component implements a level of noise within the visual content such that the altered visual content is misclassified by a trained classification model, and wherein the feature recognition component utilizes one or more feature recognition algorithms and one or more feature re-identification algorithms;

presenting an altered visual content to the practitioner within a user interface including analytics associated with the altered visual content and one or more additional recommendations, wherein the analytics and one or more recommendations are generated using one or more machine learning models;

transmitting the altered visual content to the third party following an approval by the practitioner; and

retraining the one or more machine learning models based on the one or more additional recommendations implemented and not implemented by the practitioner.

2 . The method of claim 1 , wherein altering the visual content in response to the practitioner requesting the external consultation further comprises:

displaying one or more prompts to the practitioner in the user interface, wherein the one or more prompts are designed to gather information about a downstream task, wherein adversarial algorithms, including a Fast Gradient Sign Method (FGSM), and utility metrics are leveraged in altering the visual content based on the downstream task.

3 . The method of claim 2 , wherein the altered visual content presented to the practitioner maximizes a utility of the visual content for the downstream task and minimizes a reidentification risk of the subject associated with the visual content, wherein a degree of the image perturbations, locations, and pervasiveness are throttled depending on the downstream task.

4 . The method of claim 2 , wherein the retraining of the one or more machine learning models further utilizes interactions and actions of the practitioner within the user interface during an evaluation process to fine tune the one or more machine learning models specifically to the practitioner or the downstream task.

5 . The method of claim 1 , wherein the privacy-utility system is further comprised of an evaluation component and a threshold component.

6 . The method of claim 1 , wherein the approval by the practitioner is received in the user interface following an evaluation by the practitioner of the altered visual content in an interactive environment within the user interface.

7 . The method of claim 6 , wherein actions associated with the evaluation by the practitioner are utilized for additional training of the one or more machine learning models, such that the one or more machine learning models improve future alterations to new visual content in a manner specific to a downstream task or the practitioner.

8 . The method of claim 1 , wherein the practitioner may adjust levels of acceptable utility reduction and a privacy threshold within the user interface based on the analytics associated with the altered visual content.

9 . A computer system for visual content privatization, comprising:

one or more processors, one or more computer-readable memories, one or more computer-readable tangible storage medium, and program instructions stored on at least one of the one or more tangible storage medium for execution by at least one of the one or more processors via at least one of the one or more memories, wherein the computer system is capable of performing a method comprising:

program instructions, stored on at least one of the one or more computer-readable storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to receive visual content associated with a subject;

program instructions, stored on at least one of the one or more computer-readable storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to alter the visual content using at least one or more image perturbations or one or more adversarial patches in response to a practitioner requesting an external consultation from a third party, wherein the visual content is altered within a privacy-utility system comprised of a feature recognition component and a visual content altering component, wherein the visual content altering component implements a level of noise within the visual content such that the altered visual content is misclassified by a trained classification model, and wherein the feature recognition component utilizes one or more feature recognition algorithms and one or more feature re-identification algorithms;

program instructions, stored on at least one of the one or more computer-readable storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to present an altered visual content to the practitioner within a user interface including analytics associated with the altered visual content and one or more additional recommendations, wherein the analytics and one or more recommendations are generated using one or more machine learning models;

program instructions, stored on at least one of the one or more computer-readable storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to transmit the altered visual content to the third party following an approval by the practitioner; and

program instructions, stored on at least one of the one or more computer-readable storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to retrain the one or more machine learning models based on the one or more additional recommendations implemented and not implemented by the practitioner.

10 . The computer system of claim 9 , wherein the program instructions to alter the visual content in response to the practitioner requesting the external consultation further comprises:

program instructions, stored on at least one of the one or more computer-readable storage media for execution by at least one of the one or more processors via at least one of the one or more memories, to display one or more prompts to the practitioner in the user interface, wherein the one or more prompts are designed to gather information about a downstream task, wherein adversarial algorithms, including a Fast Gradient Sign Method (FGSM), and utility metrics are leveraged in altering the visual content based on the downstream task.

11 . The computer system of claim 10 , wherein the altered visual content presented to the practitioner maximizes a utility of the visual content for the downstream task and minimizes a reidentification risk of the subject associated with the visual content, wherein a degree of the image perturbations, locations, and pervasiveness are throttled depending on the downstream task.

12 . The computer system of claim 9 , wherein the privacy-utility system is further comprised of an evaluation component and a threshold component.

13 . The computer system of claim 9 , wherein the approval by the practitioner is received in the user interface following an evaluation by the practitioner of the altered visual content in an interactive environment within the user interface.

14 . The computer system of claim 13 , wherein actions associated with the evaluation by the practitioner are utilized for additional training of the one or more machine learning models, such that the one or more machine learning models improve future alterations to new visual content in a manner specific to a downstream task or the practitioner.

15 . A computer program product for visual content privatization, comprising:

one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions comprising:

program instructions, stored on at least one of the one or more computer-readable storage media, to receive visual content associated with a subject;

program instructions, stored on at least one of the one or more computer-readable storage media, to alter the visual content using at least one or more image perturbations or one or more adversarial patches in response to a practitioner requesting an external consultation from a third party, wherein the visual content is altered within a privacy-utility system comprised of a feature recognition component and a visual content altering component, wherein the visual content altering component implements a level of noise within the visual content such that the altered visual content is misclassified by a trained classification model, and wherein the feature recognition component utilizes one or more feature recognition algorithms and one or more feature re-identification algorithms;

program instructions, stored on at least one of the one or more computer-readable storage media, to present an altered visual content to the practitioner within a user interface including analytics associated with the altered visual content and one or more additional recommendations, wherein the analytics and one or more recommendations are generated using one or more machine learning models;

program instructions, stored on at least one of the one or more computer-readable storage media, to transmit the altered visual content to the third party following an approval by the practitioner; and

program instructions, stored on at least one of the one or more computer-readable storage media, to retrain the one or more machine learning models based on the one or more additional recommendations implemented and not implemented by the practitioner.

16 . The computer program product of claim 15 , wherein the program instructions to alter the visual content in response to the practitioner requesting the external consultation further comprises:

program instructions, stored on at least one of the one or more computer-readable storage media, to display one or more prompts to the practitioner in the user interface, wherein the one or more prompts are designed to gather information about a downstream task, wherein adversarial algorithms, including a Fast Gradient Sign Method (FGSM), and utility metrics are leveraged in altering the visual content based on the downstream task.

17 . The computer program product of claim 16 , wherein the altered visual content presented to the practitioner maximizes a utility of the visual content for the downstream task and minimizes a reidentification risk of the subject associated with the visual content, wherein a degree of the image perturbations, locations, and pervasiveness are throttled depending on the downstream task.

18 . The computer program product of claim 15 , wherein the privacy-utility system is further comprised of an evaluation component and a threshold component.

19 . The computer program product of claim 15 , wherein the approval by the practitioner is received in the user interface following an evaluation by the practitioner of the altered visual content in an interactive environment within the user interface.

Assignments (1)

Continuity (1)

References Cited (48)