Open-source 'Proof of Capture' unveiled to counter Apple Reference Image, proving image authenticity via steganography

Criticizing the limitations of the 'Apple Reference Image' method released by Apple to distinguish between AI-generated and real photographed images, an open-source hardware project 'Proof of Capture' has been released as an alternative.

Limitations of Post-hoc Detection and Proving Authenticity at the Time of Capture

Given the rapid advancement of generative AI technology, 'post-hoc detection' methods that determine authenticity after image generation have fundamental limitations. Accordingly, the project approaches the problem by cryptographically proving authenticity at the time of capture. Existing methods suffer from issues where EXIF metadata is removed during image sharing or signatures are lost during editing, but Proof of Capture uses steganography to embed a signed perceptual hash directly into the pixels themselves, rather than relying on metadata.

Technical Implementation and Hardware Configuration

The initial version hid SHA-256 hashes in the least significant bits of pixels but was vulnerable to JPEG re-compression. The current version has been improved by distributing watermarks across the DWT (Discrete Wavelet Transform) and DCT (Discrete Cosine Transform) frequency domains, allowing it to withstand compression levels comparable to WhatsApp and detect content edits.

The hardware consists of a Raspberry Pi Zero, a display board, and an ATECC608 cryptographic chip responsible for generating and storing signatures. The ATECC608 chip generates private keys internally, blocks external leakage, and provides security features that lock the chip upon tampering attempts. The total production cost is under $100.

Differences from Apple's Approach and Remaining Challenges

Apple's Reference Image generates a 'digital negative' via Private Cloud Compute, but it faces criticism for keeping the verification process private and not adopting C2PA, the existing industry standard used by companies like Nikon and Sony, thereby isolating the source of trust within the Apple ecosystem. Proof of Capture has been released as open-source to break away from this closed nature. However, it still has the limitation of not providing a complete defense against 'screen attacks', where AI-generated images are displayed on a screen and photographed by a camera.