Research experiments: adversarial examples based on painted faces.

Research experiments: adversarial examples based on painted faces - 1

30 June 2020

↑ An example of a face painted in a kind of enmeshed manner (this obfuscates a few facial recognition systems including Amazon Rekognition).

↑ An attempt of applying a content-aware on Photoshop, to remove the paint: reverse the face.

↑ A result of the attempt.

↑ Loading the photoshopped face in the AWS facial analysis (the detection applied to the forehead, and the result differed from the readings of the original face)

↑ Comparison of the original face with the face that has been painted and photoshopped. "Not the same face"