Software developed by the researchers takes a 3-D model of an object and identifies key features that a human mind will be able to anchor to recreate it. Then the image is transformed into an array of ink blots and placed on a background of random blots that camouflage it.

During tests of the algorithm conducted by the team, 310 volunteers were shown blotted images with various levels of background noise and number of object splats. On “easy” setting, 98% of the emerging images could be recognized by at least 80% of the subjects.

However, the success rate dropped to just 74% as the researchers moved to “hard” mode, similar to what happens to common captcha systems as they become better disguised to beat the bots. It also took users twice as much time the average, or over 12 seconds, to identify the objects.

This problem can be solved by using an animated sequence instead of a still image. On a video, a moving shape is easily distinguishable, while in individual frames the recognition rates were as low as 9.6%.

"When we add motion we win on two fronts," says Niloy Mitra at the Indian Institute of Technology in Delhi, who was involved in the project. "Recognition becomes much easier for humans and much harder for bots."

The team is to present their work at the SIGGRAPH Asia conference in Yokohama, Japan, in December.