Du wirst das leider händisch machen müssen, da ImageMagick keine wirkliche Erkennung von Geichtern ermöglicht. Ich war übrigens überrascht das doch soviel gekommen sind! 👍

Danke, Mika. Ich hab jetzt mal testweise das Skript von der Facedetect Seite probiert und ein sehr gutes Foto mit sehr deutlichen Gesichtern durchgejagt. Das verpixelt leider nur sehr wenig und erkennt auch nicht jedes Gesicht:

https://i.imgur.com/q7zoR10.jpg

oh ein rassistisches skript?

sfb, es hat auch andere weiße Gesichter nicht erkannt. Ich hab jetzt nur einen Ausschnitt gezeigt. Schwer zu sagen, ob es nur beige Menschen erkennt. Müsste getestet werden.

Scheint nicht nur auf beige geeicht zu sein:

https://i.imgur.com/a1SQo22.jpg

I can't bring it to work. It can't find 'cv2'

Till, it needs certain Python dependencies. I just installed Facedetect it from the Debian repos and it took care of cv2. The upstream script needed some other version of cv2 and I wasn't going to bother with it.

@Rebeka Catalina posted some pixeled faces lately, maybe she can help - maybe she did it also by hand... I do not know...

some recomendation from my side: think about that process before taking the pics from demos I got used to it with the time and those which I do not want to miss in the collection I do by hand...

I just installed Facedetect it from the Debian repos and it took care of cv2.

Ok, I've got ubuntu on my main machine. I'll see, if I can find one with debian installed and try it there before trying to fix the problem on ubuntu (which seems to be more work then editing the pictures by hand). Thanks.

But facedetect is available in Ubuntu, no need for Debian directly :)

Ok, perhaps I just have to try this again, when I had some sleep.

Could be a while... :/

Ok, I successfully installed it and adapted the script a little:

mkdir -p blurred
for file in "$@"; do
  name=$(basename "$file")
  out="./blurred/$name"
  echo "$file -> $out"
  cp "$file" "$out"
  facedetect "$file" | while read x y w h; do
    mogrify -gravity NorthWest -region "${w}x${h}+${x}+${y}" \
      -scale '10%' -scale '1000%' "$out"
  done
done

on pictures with 1k-2k pixels side length and good quality, it works well, finding most faces and making them unrecognisable. On pictures with less pixels, it doesn't seem to find any faces, on pictures with higher resolution, the faces get pixelated a little, but stay recognisable.

Unfortunately, my pictures are not only high resolution (4kx3k) but also pritty low smartphone-camera-in-rain quality, so facedetect does not even recognize a single face.

I will play around a little more with it, but probably I will just select some of the better quality pictures with only a few faces and pixelate them by hand.

But thanks for the help.

I have some other question about all this pixling softwares: I have heared that it is possible to unpixel the pixeled faces if done by some software automatically. has anybody heared so too?

I have heard it, but think it can only partially be true. You can only revert it, if the information is still stored somewhere. So it depends on the method you use, when pixelating.

Of course, if you know the method, that was used, you can revert it, to all possible source states and look, which ones of them resemble a face. (Like reconstructing hashed phone numbers in Threema, Signal, etc.) But if you randomise the method enough, it shouldn't be possible.

Also, a pixelated picture that can be unpixelated with some effort, is still better than a picture that already is unpixelated.