What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the vulnerability to low - budget reverse attacks faced by Perceptual Hash when detecting and removing image - based sexual abuse materials (IBSA, such as child sexual abuse materials CSAM). Specifically, the research reveals that generative adversarial networks (GANs) can use consumer - level hardware to conduct reverse attacks on four mainstream Perceptual Hash functions (aHash, PDQ, NeuralHash, and PhotoDNA), thereby reconstructing the original image or its approximation. This not only endangers user privacy, especially for vulnerable groups, but may also lead to the leakage of sensitive information. ### Core Problems of the Paper 1. **Security of Perceptual Hash**: When Perceptual Hash is used to detect and remove IBSA materials, there is a risk of being reverse - attacked, especially low - budget attacks based on generative AI. 2. **User Privacy Protection**: Existing Perceptual Hash techniques cannot fully protect user privacy, especially when these hash values are obtained by malicious service providers or attackers. 3. **Need for Technological Improvement**: More secure hash - matching methods are required to prevent reverse attacks and ensure that the Perceptual Hash values submitted by users do not leak sensitive information. ### Specific Problems - **Vulnerability of Perceptual Hash**: Research shows that commonly used Perceptual Hash functions (such as aHash, PDQ, NeuralHash, and PhotoDNA) are vulnerable to reverse attacks, and attackers can reconstruct the original image from the hash values through techniques such as GAN. - **Privacy Risks**: Once the Perceptual Hash values are maliciously obtained, attackers can use these hash values to reconstruct the original sensitive images, thereby violating user privacy. - **Legal and Technological Challenges**: As more and more online platforms adopt Perceptual Hash techniques to detect and remove illegal content, how to ensure the security and privacy of these techniques has become an urgent problem to be solved. ### Suggested Solutions The paper proposes that more secure hash - matching methods should be implemented, such as protecting user privacy through cryptographic protocols such as Private Set Intersection (PSI), and proposes alternative solutions to implement privacy - protected IBSA content removal tools. ### Conclusion Although Perceptual Hash is effective in detecting and removing IBSA materials, the current design and implementation methods have security risks, especially in terms of privacy protection. Therefore, further research and development of more secure techniques are required to ensure that user privacy is not violated.

Perceptual Hash Inversion Attacks on Image-Based Sexual Abuse Removal Tools

Evade Deep Image Retrieval by Stashing Private Images in the Hash Space.

You See What I Want You to See: Exploring Targeted Black-Box Transferability Attack for Hash-based Image Retrieval Systems

Secure and robust image perceptual hashing

Learning to Break Deep Perceptual Hashing: The Use Case NeuralHash

Robustness of Practical Perceptual Hashing Algorithms to Hash-Evasion and Hash-Inversion Attacks

Re-purposing Perceptual Hashing based Client Side Scanning for Physical Surveillance

Deep perceptual hashing algorithms with hidden dual purpose: when client-side scanning does facial recognition

An Overview of Perceptual Hashing

Recent Development of Perceptual Image Hashing

A Novel Block-DCT and PCA Based Image Perceptual Hashing Algorithm

RevHashNet: Perceptually De-Hashing Real-Valued Image Hashes for Similarity Retrieval.

Advocating Pixel-Level Authentication of Camera-Captured Images

Deep Perceptual Hash Based on Hash Center for Image Copyright Protection

Concept Arithmetics for Circumventing Concept Inhibition in Diffusion Models

Image-Hashing-Based Anomaly Detection for Privacy-Preserving Online Proctoring

Hamming Distributions of Popular Perceptual Hashing Techniques

A replay attack detection scheme based on perceptual image hashing

Perceptual Mapping-Based Image Tamper Detection and Recovery Algorithm

Hierarchical Perceptual Noise Injection for Social Media Fingerprint Privacy Protection

Artificial Image Tampering Distorts Spatial Distribution of Texture Landmarks and Quality Characteristics