DeepSignature: Neural Watermarks for Image Authentication
A new method called DeepSignature integrates digital signatures with deep neural networks to embed cryptographically verifiable watermarks into images. These watermarks encode content and enable source attribution and integrity validation. The system is compatible with existing image formats and supports client-side verification using only the signer's public key. A novel latent-space verification approach is also introduced to detect tampering. The approach addresses threats from AI-generated images that falsely appear to originate from trusted sources.
Key facts
- DeepSignature uses neural networks to generate and embed content-encoding watermarks.
- Watermarks are cryptographically verifiable for source attribution and integrity validation.
- Compatible with existing image formats and requires no special handling.
- Supports client-side verification with only the signer's public key.
- Introduces a latent-space verification approach to detect tampering.
- Addresses threats from AI-generated images that falsely appear from trusted sources.
- Published on arXiv with ID 2604.23016.
Entities
—