Image Redaction Software

Images are redacted either automatically from detected objects or by drawing regions directly. The detection classes are the full object set: people, vehicles, license plates, weapons, devices and screens, identity documents, signatures, tattoos, street signs.

Identity documents get their own class. An ID card, driving license, passport or name tag held up to a camera or lying on a desk discloses everything on it at once. Detecting the document is how it gets redacted.

Text in an image is recognized through OCR and redacted like any other text, including in non-Latin scripts.

360-degree and panoramic imagery is handled properly: detection converts between the projection and spherical coordinates, so a detection is placed correctly rather than distorted by the projection. A tool that treats an equirectangular image as a flat rectangle puts the box in the wrong place.

The platform carries 315 distinct extensions across five content types, and images are the largest group at 187.

That number is doing real work. The awkward formats are the ones that arrive from a legacy camera system, a scientific instrument or a twenty-year-old archive, and they are exactly the ones a tool that supports "JPG, PNG, TIFF, BMP and more" will refuse. The long tail is rendered through ImageMagick, so an obscure raster format is reviewable in the browser without the software that created it.

DICOM medical imaging is handled separately, as imaging rather than as a generic image. It has its own page.

Blur, pixelate, or solid fill, chosen per job — the same three available on video. Greyscale export is available where color itself is a distinguishing characteristic.

Hundreds of images redact at once as queued work. Bulk is permissioned per format, so bulk image redaction is a separate right from bulk video. Where images share a layout — a recurring form, a standard ID scan — a template applies the same pattern across the batch.

• Face redaction · License plates
• FOIA and public records — booking photos and scene images
• Retail · Insurance claims

• Detection quality follows source quality. A low-resolution or poorly lit image detects worse than a clear one.
• Support means ingest, store and serve. Transcoding, preview and AI eligibility are stated on the capability that performs each operation, not inherited from the format count.
• Upload types are governed per portal by an accepted list and a restricted-extension blocklist, so a given deployment may accept fewer types than the platform supports.

• Still images redact with the same detection and the same three masking styles as video.
• 187 image formats, out of 315 extensions across all content types.
• Identity documents — ID cards, driving licenses, passports, name tags — are a detected class in their own right.
• The long tail renders through ImageMagick, so obscure formats preview without the originating application.
• Bulk image redaction is a distinct permission from bulk video.