Recoverable visual encryption method for human body privacy area in image collected by unmanned aerial vehicle

By using a joint deep learning model for adaptive visual encryption and decryption of UAV aerial images, the problems of privacy protection, data recovery, and real-time performance in UAV aerial image processing are solved, achieving high-fidelity privacy protection and compatibility with downstream tasks.

CN121902180AActive Publication Date: 2026-04-21UNIV OF JINAN
6 Cites 2 Cited by

Patent Information

Application Number
CN202610343632.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-20
Publication Date
2026-04-21
Estimated Expiration
2046-03-20

AI Technical Summary

Technical Problem

Existing drone aerial image processing technologies have shortcomings in terms of privacy protection, data recovery, real-time performance, and compatibility with downstream tasks, and cannot meet the comprehensive needs of drone aerial photography scenarios for accurate protection, efficient processing, and authorized restoration of human privacy areas.

Method used

By employing a joint deep learning model that combines object detection, visual encryption, and decryption modules, and through end-to-end training using adaptive pixelation transformation and composite loss functions, reversible visual encryption and high-fidelity recovery of human privacy regions can be achieved.

Benefits of technology

It achieves high-fidelity and recoverable privacy protection, adaptive dynamic adaptation, excellent real-time performance, security and reliability, non-destructive detection accuracy, and strong compatibility with downstream tasks, meeting the needs of dynamic drone scenarios.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to a recoverable visual encryption method for a human body privacy area in an image collected by an unmanned aerial vehicle. The method comprises the following steps: constructing a joint deep learning model integrating a pre-training target detection module, a self-adaptive visual encryption module and an enhanced visual decryption module; in the training stage, target detection module parameters are frozen, and an encryption and decryption module is optimized end to end through combined loss formed by a composite decryption loss function and an encryption strength loss function; during application, the target detection module accurately locates a human body sensitive area, the encryption module applies size-adaptive and strength-learnable pixelated transformation to generate a natural encryption image, and the decryption module realizes high-fidelity original information recovery through an enhanced U-Net network under an authorization condition. Compared with an existing traditional image encryption or irreversible desensitization technology, the method has the advantages that seamless integration with a mainstream target detection process is achieved, the integrity and restorability of information are guaranteed while high visual concealment is achieved, and the method has the advantages of being intelligent, high in efficiency and high in practicability.
Need to check novelty before this filing date? Find Prior Art