Quote: "if anything, thats even easier for a machine to read then the usual distorted text."
As Benjamin said, it would be impossible for an automated system to know which part of the image you were talking about. Take, for example, this image:
The system could ask any of these questions:
What is the colour of the car?
What is the colour of the house?
What is the colour of the dog?
Which item is red (car, house or dog)?
Which item is yellow (car, house or dog)?
Which item is blue (car, house or dog)?
If it was answered incorrectly you would get a new image, which would be comprised of random items in random colours.
Chart data provided with kind permission from
ELSPA