That's a good catch. This study uses images for reidentification. I wonder which other factors, not contained in an image could be used for reidentifying someone and by how much that would increase the reidentification accuracy.
If you have a publicly available photo of yourself (e.g. public social media), it's possible to match that to your genomic sequence. I.e. you already have to be anonymous online to have a chance that a private company cannot reidentify you.