Professor
Department of Psychological Sciences
School of Electrical and Computer Engineering (by courtesy)
Purdue University
West Lafayette, IN 47907-2081 >> city map
email: pizlo at psych dot purdue dot edu
Tel. (765) 494 6930
Fax: (765) 496 1264
Room #: PRCE 194

Perception is an inference that provides the observer with a veridical representation of the external world. Illusions almost never happen in everyday life. Perception is based on algorithms, not on look-up tables. These algorithms are automatically applied to the incoming sensory stimulation. This automaticity gives us the impression that perception is easy and effortless. Computationally, it is not. Perception algorithms are innate – we are born with them. It is possible, perhaps even likely that some parameters of these algorithms are estimated or tuned after birth. This makes sense because the genes that convey the information about these algorithms are communication channels with finite information capacity. The parameters may have to be tuned to varying input as the child grows up and ages. It is very unlikely, however, that the algorithms, themselves, are either learned or modified through learning during life. This permanence and innateness allows all human beings to perceive the external world the same way from their first to last day. This is critical for successful interaction among humans and with objects in the world “out there”. The key to understanding how perception algorithms work is to recognize the critical role a priori simplicity constraints play in perception. The sensory data, which comes in is always ambiguous, and it is subjected to the simplest possible interpretation. It so happens that the simplest interpretation is almost always the correct interpretation. This predilection for choosing the simplest interpretation implies that the human mind is solving an “optimization problem”. Unlike the numerical methods most scientists use for solving optimization problems with conventional computers, the human mind uses physiological mechanisms that probably implement what is called the “least action” principle of physics (Lanczos, 1970). The concept called “optimization”, which is widely used in explaining physical and biological systems, as well as in designing engineering systems, is becoming popular in explaining cognitive systems where it is likely to shed light on the relationship between the mind and the brain. It may also demystify the perennial question about whether machines can see and think.

To see how our robot solves figure-ground organization problem and performs simple navigations tasks go to Yunfeng's web site and to the VSS demo site.

Book
3D Shape: its unique place in visual perception, MIT Press.

errata

Read more >>