Verification Protocol
Technical specification of the forensic analysis pipeline. Each submission undergoes four independent checks. A photo passes verification only when all checks succeed.
Overview
Multi-Check Architecture
The verification system operates on a fail-safe principle. Rather than relying on a single detection method, multiple independent analyses must all return positive results. This reduces false positives and increases confidence in verification outcomes. Each check examines a different aspect of the image data, making it difficult to bypass the system through targeted manipulation.
Processing
Analysis Pipeline
Files are processed through a sequential pipeline. RAW data is never stored permanently. Processing occurs in isolated memory with strict access controls.
File Ingestion
RAW and JPEG files are uploaded via encrypted connection. File integrity is verified using SHA-512 hashing. Both hashes are recorded in the verification report for future reference.
RAW Decoding
The RAW file is decoded using manufacturer-specific demosaicing algorithms. A reference JPEG is generated from the sensor data using standard color science transformations.
Geometric Alignment
The submitted JPEG is aligned to the RAW-derived reference using feature matching. This accounts for cropping, rotation, and aspect ratio changes that may have occurred during editing.
Verification Checks
Four independent checks are executed in parallel: sensor authenticity, visual consistency, recapture detection, and metadata validation. Each check produces a binary pass/fail result with a confidence score.
Result Aggregation
Individual check results are combined. The overall verification passes only if all checks succeed. A detailed report is generated with per-check breakdown and confidence metrics.
C2PA Signing
For verified images, a C2PA manifest is generated and cryptographically signed. The manifest embeds provenance data directly in the image file, creating a tamper-evident record.
Sensor Authenticity
Verifies that the photo originates from a genuine camera sensor. Every camera sensor produces unique physical signatures in the images it captures. These patterns are inherent to the photographic process and difficult to replicate artificially. The check examines noise characteristics, signal distribution, and sensor-specific artifacts.
Visual Consistency
Compares the submitted photo against a reference version generated from the camera's RAW sensor data. This verifies that the submitted image accurately represents what the camera captured. The comparison uses structural similarity metrics and perceptual hashing to detect deviations.
Recapture Detection
Detects whether a photo is an original capture or a photo of a photo. Images photographed from screens, monitors, or prints exhibit telltale artifacts from the display or print medium. The check analyzes frequency domain patterns and moiré artifacts.
Metadata Validation
Examines embedded information within the photo file for consistency and integrity. Camera-generated metadata follows specific patterns and conventions that help confirm a photo's origin. The check compares EXIF fields between RAW and JPEG to detect tampering.
Sample Reports
Example Verifications
Review completed verification reports to understand the output format and level of detail provided. Each report includes the full check breakdown, confidence scores, and visual comparison tools.
Compatibility
Supported RAW Formats
The system supports RAW formats from major camera manufacturers. Only files containing mosaic Bayer pattern sensor data are supported. Demosaiced TIFF exports, Foveon sensors, and other non-Bayer architectures are not compatible.
| Format | Manufacturer |
|---|---|
| DNG | Adobe (Universal) |
| CR2 | Canon |
| CR3 | Canon (newer models) |
| NEF | Nikon |
| ARW | Sony |
| RAF | Fujifilm |
| ORF | Olympus / OM System |
| RW2 | Panasonic |
| PEF | Pentax |
Input Requirements
Submission Format
Verification requires both the camera RAW file and the published JPEG. The RAW file provides the ground truth reference. The JPEG is the asset under examination.
RAW File
ReferenceOriginal sensor data from the camera. Must be the unmodified file as written by the camera.
Embedded previews are ignored. Only sensor data is used for comparison.
JPEG File
TargetThe published or distributed version of the photo. This is the asset being verified against the RAW source.
Standard JPEG compression is expected. Progressive encoding is supported.
Output
Verification Results
Each verification produces a detailed report with per-check results, confidence scores, and visual comparison tools. Verified images receive a C2PA manifest for downstream verification.
Verification Report
A structured record of the verification including file hashes, individual check results, and overall status. Reports can be shared via unique URL or downloaded as PDF.
C2PA Manifest
Coalition for Content Provenance and Authenticity manifest embedded in the verified image. Contains cryptographic signatures binding the verification to the image data.
Conforms to C2PA specification version 1.3. Signed using ECDSA P-384.
Manifests can be verified at verify.contentauthenticity.org or using CAI browser extensions.
Editing Tolerances
Acceptable Transformations
The system accounts for standard photographic post-processing. Certain transformations are expected and will not cause verification to fail. Destructive edits that alter image content will be detected.
Permitted Operations
- •Exposure, contrast, and white balance adjustments
- •Cropping and aspect ratio changes
- •Standard sharpening and noise reduction
- •Color grading and tone curve adjustments
Operations That Will Fail Verification
- •Content-aware fill, object removal, or cloning
- •Compositing elements from multiple sources
- •AI-based upscaling or generation
Run Verification
Submit a RAW and JPEG pair for forensic analysis. Results are available immediately.