This thread is about those little boxes that have words that you have to read and type into a new box in order to do things like register for websites. The letters are usually wavy, or crisscrossed, and inside of a grainy area. What’s the purpose of this practice?
Are there programs out there that can “read” regular text and copy it into the little box?
Just a WAG, but I guess it would possible for a cleverly written robot to compare the picture to a set of character designs, checking for matches. Essentially this would mean the program could “read” the box. The measures mentioned in the OP might make that harder.
Absolutely. In fact, there are computer programs that can read the squiggly words, too.
The squiggly words are one type of CAPTCHA test (basically, a test that one computer can generate on the fly and know the answer to, that a human can figure out the answer to, but that another computer can’t figure out the answer.
There are lots of different ways to do image processing on something if you know basically what to look for. OCR (Optical Character Recognition) programs are getting quite good at “reading” undistorted text.
Some of the techniques are completely useless. For example, the colored backgrounds. The computer can just filter that out. Even when the text is distorted, you can still make a pretty good guess. It might confuse a “g” with a “q”, because both are basically circles with tails, but it is unlikely to mistake either one for an “x” or an “f” (for example). When you add in the fact that many CAPTCHAs use dictionary words, you solve even more potential problems distinguishing between letters.
Here’s a page where some Berkeley researchers explain a bit about how to break some of the visual CAPTCHAs. They are not 100% successful, but since these tests are often employed to keep spammers from getting lots of free accounts, even a fractional success rate is plenty when they can submit many attempts every minute.
The goal is to make it so that humans can register, but spam bots can’t.
I forgot to mention about a few of the other techniques you mentioned:
The grainy or crisscrossings are used to break up the outline of the letters. If the computer can find the “edges” in the image, it’s relatively easy to get the general shape of the letters. Of course, criss-crosses can be filtered out because they are a regular pattern, and grainy images can be “smoothed” until the edges are more obvious, but both introduce some more uncertainty.