A sample isolation mechanism-based method and system for pedestrian re-identification to protect identity privacy.

By employing an identity protection method based on a sample isolation mechanism, the generator produces the optimal perturbation image, solving the problem of overlapping identities in pedestrian re-identification, achieving effective privacy protection for the target person, and improving the success rate of identity protection and image quality.

CN115641609BActive Publication Date: 2026-07-17INSTITUTE OF INFORMATION ENGINEERING CHINESE ACADEMY OF SCIENCES
View PDF 1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
INSTITUTE OF INFORMATION ENGINEERING CHINESE ACADEMY OF SCIENCES
Filing Date
2022-10-11
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing pedestrian re-identification technologies have vulnerabilities in protecting the privacy of the target person's identity. Malicious users can leak personal identity information through adversarial attacks, and existing adversarial image protection methods are prone to overlapping identity issues, resulting in insufficient protection capabilities.

Method used

An identity protection method based on sample isolation mechanism is adopted. Adversarial images are generated by a generator, and the generator is trained using identity isolation loss, misclassification loss and misordering loss to generate the optimal perturbation, so that the adversarial image is far away from the original identity and other identities, forming an isolated position to defend against malicious queries.

Benefits of technology

It effectively protects target individuals from unauthorized queries and searches, improving the success rate of identity protection. It is particularly superior in the face of random and sequential attacks, and the image quality remains high.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115641609B_ABST
    Figure CN115641609B_ABST
Patent Text Reader

Abstract

This invention relates to an identity privacy protection method and system based on a sample isolation mechanism for pedestrian re-identification. The method includes: given a clean image, a generator learns a perturbation to generate a corresponding adversarial image; inputting the adversarial image and the clean image into a target model to obtain adversarial features and clean features respectively; training the generator on the target model using the adversarial features and clean features, employing identity isolation loss, misclassification loss, and misranking loss to generate the optimal perturbation; adding the generated optimal perturbation to the image of the target person to obtain the adversarial image of the target person; and protecting the target person from malicious queries through the adversarial image. This invention proposes an identity isolation mechanism that explicitly restricts the adversarial image to an isolated location, ensuring that the target person is far removed from both their original identity and any other identity, thereby protecting them from illegal query retrieval.
Need to check novelty before this filing date? Find Prior Art