Azure Media Analytics Face Redaction Screenshot

Microsoft’s Azure Media Analytics is  a collection of speech and vision components that allow organizations to create actionable insights from their video files. However, with all that data processing comes some privacy concerns.

In public safety and news media scenarios, it’s often difficult to protect the identity of individuals in footage and can take many hours of post-processing.

According to program manager of Azure Media Analytics Richard Li, Microsoft has come up with a solution to this problem.

Face Redaction

Microsoft’s face redaction technology combs through each video frame, tracking and blurring faces as they move. Thanks to the smart tracking, individuals can be blurred regardless of their angle.

However, Li warns that the service isn’t 100% accurate. A real person will always get better results, especially in low light or high movement scenes.

As such, Azure Media Analytics allows for two modes. The first is fully automatic, whereas the second is a “two pass workflow.” Customers can use this to make frame-by-frame adjustments, as well as select and deselect faces dependent on IDs.

The results are pretty impressive, and the service supports up to 64 human faces per frame. However, according to Li, there are still some limitations:

  • “The supported input video formats include MP4, MOV, and WMV.
  • The detectable face size range is 24×24 to 2048×2048 pixels. The faces out of this range will not be detected.
  • For each video, the maximum number of faces returned is 64.
  • Some faces may not be detected due to technical challenges; e.g. very large face angles (head-pose), and large occlusion. Frontal and near-frontal faces have the best results.”

To start with face detection, you simply have to create a Media Services account within your Azure subscription. You can then use the REST API/SDKs to gain access, or the Azure Media Services Explorer.