午夜福利1000集合

Picture puzzles separate human from machine

A tool for hiding images within larger pictures could provide a neat way of distinguishing between people and bots online


Video: Camouflaged images

Can you spot the hidden animals?
Can you spot the hidden animals?
(Image: Hung-Kuochu)
Picture puzzles separate human from machine
(Image: Hung-Kuochu)
Picture puzzles separate human from machine
(Image: Hung-Kuochu)

A way to create children鈥檚 picture puzzles could be used to improve online authentication, as it involves spotting images within images.

There are four animals hidden in the accompanying picture (and you can find the solution here). To obscure them, of National Cheng Kung University in Tainan, Taiwan, developed a tool that first simplifies the animal image, creating a pattern of lighter and darker regions. The image is then overlaid on the base picture so that the background texture shows through. Any gaps are filled with texture that approximates the background. 鈥淥ur model can generate camouflage images at different difficulty levels,鈥 Chu says.

The technique works best with natural backgrounds that have plenty of detail, such as cliff faces, forests and cloudy skies, Chu says. Large blocks of solid colour, or regular patterns like those of brick walls, are more difficult to hide images in.

Though it is not always easy for humans to pick out the images, we can do it. However, it is almost impossible for computers, Chu says. That means as well as generating photo puzzles, the system could be used to make Captcha images, which websites employ to ensure that users signing up for accounts are genuinely people, rather than software bots.

鈥淭he work really is creative and their graphic tool is very neat,鈥 says , a psychologist at the University of California at San Diego.

Chu鈥檚 work fits with the theory that the brain can only be conscious of one visual feature at a glance, but can track the locations of multiple features simultaneously, he says. As the viewer spends time looking at the image, the locations of recognised features accumulate until the object as a whole enters the viewer鈥檚 attention.

The work will be presented at the conference in Los Angeles later this month.

Topics: Computer crime