http://en.wikipedia.org/wiki/Captcha
A number of research projects have attempted (often with success) to beat visual CAPTCHAs by creating programs that contain the following functionality:
Extraction of the image from the web page.
Removal of background clutter, for example with color filters and detection of thin lines.
Segmentation, i.e. splitting the image into segments containing a single letter.
Identifying the letter for each segment.
Steps 1, 2, and 4 are easy tasks for computers. The only part where humans still outperform computers is segmentation. If the background clutter consists of shapes similar to letter shapes, and the letters are connected by this clutter, the segmentation becomes nearly impossible with current software. Hence, an effective CAPTCHA should focus on the segmentation.