|
CAPTCHA Experiment
Theory This CAPTCHA method is an Experiment. Due to tools like CURL, hacking CAPTCHAs is possible by
processing the image using algorithims that look for contrast, and edge detection. This mehtod
uses equal amounts of contrast, and plenty of false edges. We also use multiple images.
If both images are aligned and combined properly, the intended text should appear. However if the images
are not aligned correctly, or processed individually, the intended text should be impossible to parse.
However this relies on the browser's ability to display transparencies in images, and align images
using CSS. Browsers like CURL do not render content, so the CAPTCH hacking program would have to
render the content, and then search for the captcha.
Experiment On the left are the source images. The top right should be a captcha with a single word sticking out.
The bottom right should be a cpatcha with no words visible because it has been misaligned.


Conclusions Failure: A human still has a hard time reading the words.
Further Work Make it readable.
Some problems that occoured while creating the image, were that this
particular implementation uses 50% opacity on the top layer. So, if a
random pixel set was used around the password, a large grey area appeared around the text.
However, the text is the only thing that should be grey. To solve this we had
to copy the bottom layer's pixels. Which you will see as the top-most layer in the
photoshop file.
This creates a problem, because the image could be aligned programatically
by searching for a patterns in the colors ignoring the alpha channel.
One solution is to use variable opacity on the password, and solid on the garbage,
however this would leave the password exposed by removing all solid pixels. So
a random number of solid and opaque pixels could be distributed along the image in
further experiments. Although I think this would still leave the password difficult to read.
- Robert Wallis
November 8, 2006
|
|