Petfinder currently has 253,065 animals listed on it. I’m almost certain that 200,000 of those animals are cats or dogs. Let’s say that half of those have pictures (it’s probably more–most groups use pictures these days). It’s a busy spammer who is going to go through and identify all those pictures.
And next week, they’ll have to do it again: Petfinder draws from animal shelters, where animals are constantly cycling in and out.
If image recognition software becomes good enough that a computer can figure out that this is a cat and this is a dog, then this system will stop working. But I’m not aware of software that is that good yet.
I would say it’s quite a bit better, actually. All it has to do is identify the eyes and nose. Once you do that, it becomes immediately obvious by the ratio of the distance between eyes compare to the distance from the midpoint between the eyes to the nose.
Look at those pictures again. How hard would it be to identify the eyes and nose?
That is an interesting point, but it seems like it is a bit more difficult than that. The camera’s orientation relative to the subject would change the perceieved distance between the eyes and nose. For example, a picture of a dog taken directly in front of the dog’s nose would make the distance appear closer than a picture taken from above the dog’s head.
Likewise, the dog example up-thread shows another pair of complications. The second and third pictures show two dogs and one of the dogs is facing sideways. So the picture has exactly three dog eyes. A similar effect could be generated dynamically by croping the picture randomly on the fly.
There is a partnership between Petfinder and Microsoft called Asirra. Images on petfinder are already classified by species, not just limited to dogs and cats. If you include adopted animals, there are well over a million images from which to select. Additionally, each image has a link to the animal’s details which will hopefully help to raise awareness.
I am not sure how many web sites have implemented this or how well it has worked as a CAPTCHA or if it has increased adoptions.
Thanks for the link! Trying it out on that site makes it clear that Ellis Dee’s method wouldn’t work:
-Some of the animals are in profile
-Some of the animals have closed eyes
-Some of the calico cats have weird patterns that could frustrate computer identification of their eyes
-Some of the pictures have multiple animals
-Some of the pictures have multiple pictures (i.e., they’re sort of a collage).
I once saw a site use a captcha based on numbers:
You had to type in the answer to the math problem shown in the captcha window
the problem was something like :
“346 plus one equals” (this was shown as a gif graphic or something similar, easy to read for a person, and I presume not so easy for a computer)
this seems like a perfect system to me: the human user only has to type 3 digits, and the bot presumably can’t read the question.
Am I missing something here?
A whole bunch of these potential CAPTCHAs are actually more limited than the existing ones because they either have an existing key that just has to be mined or are only good for small sites.
For a CAPTCHA to be good, it has to be useful to anyone. Any solution where each site has to make up its own questions or design its own solution that’s obscure enough that no one will bother to figure it out is not particularly helpful. It might be an ok solution for a few small sites, and there’s nothing wrong with that, but it will require a lot of reproduced work and won’t solve the general problem, since it won’t solve the problem for any site big enough to be a major target on its own.
For all the ones using pictures, the big problem is that the stash of pictures is known and public. Most of them are even tagged by users. The bots don’t have to use image recognition to figure it out, they can just get a site dump of Petfinder or Flickr or whatever and associate tags and other meta information with the image. Furthermore, given enough time, the bot will learn from its mistakes. Even though some of the photos it has are likely to be mislabeled, it can figure that out pretty quickly if it’s allowed repeated access to the authentication procedure.
The only accurate solution I can think of is a retina scan accompanied by a photocopy of their social security and a valid driver’s license. And a DNA test.
Again, Petfinder has literally hundreds of thousands of images, and these images rotate in and out on an hourly basis as animals at shelters are adopted/returned to owners/euthanized. IT would be extremely difficult for a user to tag all the images, or for a bot to learn all the images.
Petfinder gave the Microsoft developers access to their entire database, not just the public images - about 3 million images and growing at about 10,000 per day. Only about 10% of the database represents animals for adoption and therefore publicly available. Additionally, the bots would not just have to guess right on one photo, Asirra presents 12 images where the viewer needs to identify a specific animal.
Microsoft also limits the number of challenges in a day, over 500 failed HIP challenges will result in automatic failure for several hours and also alerts network admins.
Granted, this is all based on information from the Microsoft web site , so I am sure that there will be holes discovered later.
Best solution: find at least ONE of the spammers responsible for, well, SPAM, and publicly torture and kill them, Inquisition-style, and put out notice that we WILL find other spammers and they WILL suffer the same fate. And maybe to drive the message home even more we invade a small country, declare that the name of the country is changed to “Spam” and that the inhabitants are “Spammers” then we send in troops to rape, pillage, and kill the entire country on live TV, then have a stern voice-over say “Are YOU a spammer? THIS is what will happen to your family, friends, everyone you ever cared about. And it’s just the BEGINNING of what will happen to you. Stop now and repent, or you will find yourself praying for a death that will not come.” Then put one of those disclaimers at the bottom of the screen, you know, the ones that they put on car commercials where a truck gets attacked and spat out by the Loch Ness Monster and it says “DRAMATIZATION” just in case someone tries to complain about the announcement. Yeah, that’ll work.