Image Safe Search — Sensitive Content Detection

Module 6 · Image Safety

Image Safe Search — Sensitive Content Detection

Ensure scanned documents and uploaded photos do not contain sensitive or inappropriate content, using configurable policy categories and privacy-preserving processing.

← Modules

RFI module deep-dive

Image Safe Search — Sensitive Content Detection

Prototype page exposes input contracts, processing stages, structured outputs, quality controls, privacy controls, integration points, PoC questions and evidence needed before production acceptance.

Open PoC Benchmark Lab

Inputs

Image / scanned documentUploaded onboarding images after malware/type validation

Category policyAdult, violence, racy, medical, spoof; configurable

Threshold profilePolicy-defined confidence threshold per category

DeploymentPrivate/hybrid/on-prem processing target

Processing Pipeline

Secure decode

Validate image type, strip unnecessary metadata and create isolated processing object.

Safety model

Run multi-label content-safety classification.

Spoof / recapture signal

Optional detection for image recapture or synthetic/spoof indicators.

Thresholding

Apply policy version to category scores.

Privacy-safe evidence

Persist labels/scores; retain raw image only per retention policy.

Decision routing

Block unsafe upload, request replacement or send to review depending on policy.

Structured Outputs

category_scoresAdult / violence / racy / medical / spoof

policy_hitsCategories exceeding configured thresholds

actionPass / block upload / manual review

model_versionExact safety model/version

policy_versionThreshold configuration version

latencyMeasured per request in PoC

✓

Sensitive-content detectionDetect adult/explicit, violence and other inappropriate visual content in uploaded images.

✓

Configurable categoriesSupport categories such as adult, violence, racy, medical and spoof with configurable thresholds.

✓

Confidential processingProtect images because they may contain corporate documents or personal data.

✓

Private/on-prem optionSupport deployment model suitable for sensitive image/document processing where required.

✓

Performance metricsMeasure per-category accuracy, latency and throughput.

#	Question to provider	Prototype status	Evidence / response expected
1	Which content categories can be detected and what is the accuracy for each?	PoC response	Provide per-category precision/recall and validation-set composition.
2	Is private/on-prem processing available for sensitive images?	PoC response	Provide deployment architecture, data flow and storage behavior.
3	Can detection thresholds be configured to Finpay policy?	PoC response	Show threshold administration, versioning and change audit.
4	What image-processing latency and throughput are supported?	PoC response	Provide p50/p95/p99 and concurrent throughput under test conditions.

Sensitive image handling

Images may include IDs or other personal data; processing must be isolated, encrypted and tightly access-controlled.

Metadata stripping

Remove EXIF or other metadata that is not required for the screening purpose.

Private processing

A private/on-prem route reduces unnecessary exposure of sensitive images to third parties.

No raw-image logging

Operational logs contain identifiers and scores, not full image payloads.

Illustrative structured output

Schema is intentionally explicit to support decision-engine integration, explainability and audit. Values are simulated.

{ "module":"image_safe_search", "scores":{"adult":0.004,"violence":0.008,"racy":0.012,"medical":0.006,"spoof":0.07}, "threshold_profile":"FINPAY-KYB-IMG-v1", "policy_hits":[], "action":"pass", "processing_mode":"private", "raw_image_logged":false }

Evidence inspection

Image-safety results are retained as minimized policy evidence with threshold, model version and private-processing metadata.

Open Evidence Center →