46

Now this is what a captcha should look like...

Comments
  • 2
    Does a better job and better UX.
    @AlpineLinnix
  • 1
    More pleasant and not so easy to bypass since the icons changes and different IDs and position.
  • 10
    Bots would fail, because they would try to click themself.

    But jokes aside, I feel like this would be easy for a picture recognition tool.
  • 5
    It would be especially easy because each one of those is presumably its own div with its own image source meaning the bot wouldn't even need to parse the pic, just run all 5 of then through a classifier and use a word2vec on the label to find the label with the same meaning as the classification. Problem solved.
  • 1
    Won't be so easy as those images are getting a random ID per load. Would be easy if they were static. + the ID doesn't correlate with the supposed image.
    @hexc
  • 1
    @mokoshalb even if the div had a random Id, just use a more specific selector like .captcha > div > div or w/e. As for the random id, store that as a value, run it through the CNN and put the classification as the key like { car:"sh63d737", computer:"hs482jcu"...} once you have the map contain all the image classifications/div ids, do the word2vec on the classification, find the proper one, and do a get element by Id with the correct answer and .click() the div. DONE, EASY.
  • 0
    @mokoshalb if your bot fails because if that, you suck at bot writing
  • 0
    Where did u find it?
  • 0
    Startapp dot com
    @fastidio
  • 0
    @irene not sure, what do you mean by voodoo in this context?
  • 0
    @irene ah, in that situation, you could just use one of the various js libraries to convert the div to an image.
Add Comment