Blurring faces or license plates is often treated as a quick video fix. In practice, it’s a risk-control measure – and it only works when identification becomes impractical for both humans and automated tools throughout the entire publishing pipeline: export, re-encoding, platform compression, and common “enhancement” filters. That’s why three things matter most: full coverage (no gaps), sufficient strength, and repeatable quality control before release.
What Does “Strong Enough” Mean in Practice?
Operationally, footage is edited more safely when it meets all three conditions at the same time:
- High-sensitivity detection – it’s better to catch more and discard false positives than to miss one critical frame.
- Coverage with margin – the mask also includes auxiliary areas (hairline, beard, license plate frame).
- Resilience to the publishing pipeline – after export and compression, the object still cannot be meaningfully read or matched.
The Most Common Mistake: A Blurred Face, a Recognizable Person
Many organizations “cover” only the center of the face. That is often not enough, because recognizability is also supported by cheeks, jawline, hairstyle, ears—and in video, additionally by body shape and scene context. From a data-protection perspective, the key question is whether the material allows identification in a reasonable way—and that depends on the whole frame, not just the face pixels.
Face Settings That Usually Hold Up in Real-World Footage
1) Mask Margin
A good setup starts with geometry. In practice, it’s worth expanding the face area by about 10–30%, depending on framing and pose. Small faces in a crowd are particularly risky – if the object is very small (e.g., below about 20–24 px wide), a safer rule may be full masking rather than subtle blur.
2) Strength: Not “Light,” but Effective
Most commonly used methods include Gaussian blur, pixelation, or a solid block. As starting values for 1080p, you can assume:
- face close-ups (typically 120–200 px) – sigma ~ 12–20 or pixelation with 16–24 blocks,
- crowds (24–80 px) – similar ranges, but more frequent use of full masking in difficult shots.
For 4K, strength usually needs to be increased roughly in proportion to resolution (often about 1.8×–2× the 1080p settings), and then confirmed with post-export tests.
License Plates: The Risk of a Single Frame
Plates are tricky because one “sharp” frame can be enough for a readable capture. That’s why mask stability between frames and post-compression testing matter.
1) Margin and Inter-Frame Stabilization
Expanding the plate mask by 5–15% helps cover frames and elements that facilitate OCR. In video, it’s worth using tracking and interpolation – this minimizes flicker and short masking dropouts.
2) Starting Parameters and an OCR Test
For typical road footage in 1080p, a good starting point is:
- sigma ~ 10–16 for Gaussian blur, or
- pixelation with 12–20 blocks.
The most important QA test: run OCR on plate crops and confirm that characters are not correctly recognized – after exporting to the target format, not only in the working preview.
Quality Control That Protects You From an Embarrassing Release
If you need to choose a few steps that deliver the most impact, these typically do:
- “Worst-case” samples: night, IR, rain, reflections, fast motion, strong close-ups.
- Retest after export: the same codec, bitrate, and profile that will be published on your site or social media.
- Robustness tests: sharpening, super-resolution, denoising—and check whether face recognition and OCR still fail.
- Manual review of edge cases: crowds, occlusions, side profiles, helmets, masks.
- Documented settings: so the process is repeatable and you can demonstrate due diligence.
On-Premise Process and Repeatable Workflows With Gallio PRO
In many organizations, raw recordings cannot “leave” the infrastructure – due to security, confidentiality, and access control. A practical choice is therefore an on-premise model where editing happens locally.
If you need an on-premise photo and video anonymization tool that automatically blurs faces and license plates, Gallio PRO is worth considering. Gallio PRO does not work in streaming mode and does not perform real-time anonymization. It also does not offer automatic detection of logos, tattoos, ID badges, documents, or screens – these elements can be redacted manually in the editor.
To test settings on your own footage, you can download the free Gallio PRO demo.
FAQ – Strong Face and Plate Blurring
Can you set one “safe” parameter for all recordings?Rarely. Materials differ in resolution, framing, and context. The best approach is default parameters plus QA testing and fallback rules (e.g., a full mask for small objects).
Why can a working preview be misleading?Because publishing usually includes export and compression, which can change the visibility of details. That’s why verification should be performed on the final file.
Which is safer: pixelation or Gaussian blur?Both can be effective at the right scale. For small and problematic objects, full masking is often the most reliable option.
What’s the simplest way to check that a license plate is truly unreadable?Run OCR on plate crops from the post-export file. If OCR returns characters, the settings should be strengthened.
Does Gallio PRO also blur silhouettes or logos?No. It automatically blurs faces and license plates. Other elements require manual redaction in the editor.
References
- GDPR – Regulation (EU) 2016/679, including Article 4 and Recital 26
- EDPB – Guidelines 3/2019 on processing personal data through video devices
- ICO – What is personal data (including photos and recordings)
- ICO – CCTV and video surveillance
