It is not the addition, that is hard for computers. What is quite hard for computer vision algorithms is to count the number of dots on dices in an image.
Good. So why require the user to add up the values? Some alternatives:
* Click all the dice with value 5.
* Click all the pictures with a roll of 4 and 5. (each picture shows a pair of dice).
If you test for pattern recognizing the number of dots on the top of the dice, then you can just verify only that. No need to make a human user to do a task that a computer can easily do.
I can imagine a bunch of users pulling out their phone calculator app to do the addition, which should tell how stupid this captcha is.
