Tencent Cloud Face Static Liveness Detection Accurate

Problem

In account onboarding, payment verification, and remote account opening, simply checking whether an image contains a face is not enough to stop attackers who use high-resolution screen photos, printed paper cutouts, or 3D masks. tencentcloud-faceid-detectlivefaceaccurate targets static image scenarios and asks a more specific question: is this image from a real live face, or is it a replay, synthetic, or advanced attack artifact?

How It Works

The skill calls Tencent Cloud Face Recognition's high-accuracy static liveness detection API and turns image input into a liveness decision. Key capabilities include:
- High-resolution screen attack defense: identifies photos captured from high-resolution screens.
- Printed paper attack defense: identifies re-photographed printed photos.
- 3D mask attack defense: identifies advanced attacks such as 3D masks.
- Multiple input types: accepts a local image path, a Base64 string, or a remote image URL via --url.

In practice, you typically run scripts/main.py and choose parameters based on the scenario:
1. Pass a local file or Base64 string with --image, or pass an image URL with --url; if both are provided, the URL takes precedence.
2. Optionally set --face-model-version; the current supported version is 3.0, which is also the default.
3. Optionally set --region; if empty, the default region is used.
4. The API returns a JSON result. A higher liveness score indicates a higher likelihood of a real face; the documented recommended threshold is typically 40, so a score at or above that threshold can be treated as a real face.

Boundaries and Notes

• Supported formats are PNG, JPG, JPEG, and BMP; GIF is not supported.
• Resolution limits apply: JPG long edge must not exceed 4000 pixels, while other formats must not exceed 2000 pixels.
• A width-to-height ratio close to 3:4 is recommended, and the face region should be larger than 100×100 pixels.
• The Base64-encoded image must not exceed 5MB.
• This skill is for static image liveness judgment, not real-time video streams, multi-frame motion liveness, or complete risk-control systems that require complex business policies. In production, treat the liveness score as one signal and combine it with device, behavior, and business rules.