The Defence problem: provenance at volume
Imagery from conventional sensors, autonomous platforms, public sources, partner systems and synthetic-generation pipelines can enter the same analytical workflow. When the volume exceeds the capacity for detailed examination, origin assessment becomes partly a triage question: which images are consistent with a camera-origin process, and which warrant further scrutiny?
RHEM Labs proposes investigating SPOT for that task. This is a potential application of the laboratory’s research, not an assertion that Defence has adopted the method or identified a procurement requirement for it.
What SPOT examines
Sensor Pattern Origin Testing is an experimental method examining the statistical behaviour of noise residues associated with image formation. It asks a narrow question about image-origin consistency. It is not image-semantic classification and does not establish whether a depicted event is true.
The main SPOT page describes the scientific basis, published study, further experimental comparison and research limitations. A camera photograph of synthetic material is a particularly important distinction: a genuine camera process may capture a scene or display whose content has a different provenance.
Where SPOT could fit
- Imagery enters the workflowRetain identifiers, available provenance and processing history.
- SPOT triageApply a specified method version within an evaluated image population.
- Origin-consistency resultReport the score, tested interpretation and uncertainty; allow an inconclusive outcome.
- Analyst prioritisationCombine the result with context and the consequences of a missed or incorrect flag.
- Further examinationDirect selected material to specialist exploitation, forensic examination or other provenance methods.
This proposed workflow complements provenance metadata, cryptographic provenance, intelligence assessment, conventional image forensics, analyst judgement and other sensor information. It does not replace them. A confidence label would require calibration against the intended population; a score should not be represented as the probability that a scene is true.
Characteristics to evaluate
Speed and scale. Measure throughput, latency, memory and failure rates on representative hardware and image sizes. Include preprocessing and integration overhead. Operational throughput remains to be established.
An explainable forensic basis. Identify the measured characteristics, hypotheses, decision rule and conditions in which the interpretation may fail. Explainability requires evidence about the method, not simply a plausible account of sensor physics.
Integration potential. Evaluate how source identifiers, method versions, results and exceptions would move through an existing analytical system. Interface design, deployment environment and security assurance remain work to be agreed and tested.
Triage before detailed examination. Test whether prioritisation reduces analyst effort without an unacceptable missed-detection rate. The appropriate threshold depends on the mission, population and consequence of error.
Existing evidence and its limits
The published experiment uses Dresden camera and DALL·E imagery and reports repeated testing. The main page sets out the reported measures and discrepancies between the abstract, Results means and a single example. The separate experimental performance profile is also available there.
Those results do not establish performance on Defence imagery. Evaluation for a Defence use would require a representative dataset, a fixed implementation version, run-level variation, independently verified throughput and testing in the intended workflow.
A test-and-evaluation programme
An evaluation should start with the mission problem, relevant imagery population and required performance. Fix the design before testing held-out material, document exclusions, compare methods on the same eligible data and retain run records. Evaluate error rates at the proposed triage threshold, calibration where appropriate, abstention and the cost of sending material for further examination.
The next questions include different camera pipelines; screenshots; recapture and camera photographs of synthetic displays; recompression; partial manipulation and inpainting; degraded operational imagery; mixed provenance; adversarial conditions; and integration into wider analytical systems. Each may require a different test population rather than one pooled accuracy figure.
- Laboratory prototypeCurrent research status; define method and hypotheses.
- Representative evaluationExtend testing to relevant populations and documented operating conditions.
- Operational-user testingAssess interpretation, analyst workload and decision consequences.
- IntegrationTest interfaces, provenance retention, repeatability and systems assurance.
- Independent validationSeek scrutiny beyond the development team and resolve material failures.
- Deployment-readiness decisionAssess the evidence for a defined use, including support and limitations.
This proposed maturation path includes independent scrutiny from the outset. Iterative development uses failures to refine the method and evaluation before progressing.
Australian Defence strategy and capability development
The 2024 National Defence Strategy, paragraphs 8.6–8.9 (printed page 56), discusses strategic partnerships and minimum viable capability. The 2024 Integrated Investment Program, paragraph 1.17 (page 13) and paragraphs 1.24–1.26 (page 16), addresses test and evaluation, systems assurance and iterative capability delivery. Defence’s 2024 NDS and IIP publication page.
The Defence Innovation, Science and Technology Strategy: Accelerating Asymmetric Advantage – Delivering More, Together, pages 19 and 22–23, describes partnerships, experimentation and test and evaluation; page 51 defines minimum viable capability and spiral development. Defence’s strategy release and source link.
These strategies emphasise experimentation, partnerships and the transition from research to capability. RHEM Labs proposes testing whether Australian image-provenance research can contribute a useful component to that wider system. The starting point is a defined mission problem and an experiment capable of establishing whether the method produces a useful effect. Minimum viable capability requires evidence of that effect in context.
Discuss evaluation or capability integration
RHEM Labs welcomes discussion with Defence, national-security, government and industry partners able to contribute operational requirements, suitable test datasets, evaluation environments, integration pathways or independent validation.
Begin with mission problem → imagery population → performance requirement → evaluation design → integration pathway. This supports a deliberate transition from research to capability if the evidence warrants it.
Discuss evaluation or capability integration. Initial contact should be unclassified and non-sensitive. Handling and access arrangements must be established before any restricted material is shared. See the broader Defence & government pathway and Dr Matthews’ expertise.