TA University, Saudi researchers collaborate on computer 'bot' problem

TA University, Saudi res

By JUDY SIEGEL-ITZKOVICH
computer programmer 248.88 (photo credit: AP)
computer programmer 248.88
(photo credit: AP)
Many Internet users are clueless about the collection of distorted letters and numbers they must often identify and key in to get entrance to various Web sites. These are called Capthcha - meant to be a barrier that prevents computer automatons ("bots") from getting into sites and voting, opening financial accounts or otherwise influencing content. Now researchers at Tel Aviv University - part of an international team - have developed a "synthesis technique" to overcome the "bots" by generating images of animated 3-D objects that are detectable by humans but difficult for an automatic algorithm to recognize. The team was headed by TAU computer studies department Prof. Daniel Cohen-Or and - unusually - included colleagues at King Abdullah University in Saudi Arabia, The University of Delhi in India and researchers in Taiwan. Their findings are being presented this week at the Second ACM SIGGRAPH Asia Conference being held in Yokohama, Japan, where the world's leading experts on computer graphics, interactive techniques, art and animation are in attendance. Just a month ago, the anti-virus company AVG reported that computer hackers managed to break conventional Capthcha defenses as an exercise in an academic course. One of the main reasons is that digits and letters comprise a finite source of data, and even if they are twisted, algorithms could still be used to deal with the letters and numbers even after they were made to look malformed. Internet accounts created by the hackers were used to spread junk mail and links to Web sites with viruses, Trojan horse computer programs and other hostile codes. The TAU development is based on emergence, the unique human ability to aggregate information from seemingly meaningless pieces and perceive a whole that is meaningful. While single images look like nothing significant, the processed image seems very clear to the human mind. This special skill of humans can constitute an effective scheme to tell humans and machines apart, says Cohen-Or, who notes that a drawing consisting of squiggles can turn into a movie-like image of motion when there are slight variations from one to the next and they are flipped rapidly. The new technique allows the generation of an infinite number of images with 3-D "emergence figures," with the team's algorithm designed so that it seems as if the synthesized images divulge little useful information or cues to assist any segmentation or recognition procedure. Computer vision algorithms are incapable of effectively processing such images, he notes. "However, when a human observer is presented with an emergence image, synthesized using an object [one] is familiar with, the figure emerges when observed as a whole. We can control the difficulty level of perceiving the emergence effect through a limited set of parameters." Cohen-Or was assisted by Lior Wolf and Hezy Yeshurun of TAU, Prof. Niloy Mitra of the Technological Institute of Delhi in India, researchers at King Abdullah University for Science and Technology in Saudi Arabia and two Taiwanese researchers. Cohen-Or says greater development is needed before the new technique can be of practical use.