News Release

What We Expect Is Often What We See

Peer-Reviewed Publication

Max-Planck-Gesellschaft

Researchers at the Max Planck Institute for Biological Cybernetics in Tuebingen/Germany and the University of Wisconsin-Madison reported in nature neuroscience, Vol. 1, No. 3, July 1998 new experimental findings about the interaction between depth perception and object recognition and visual processing in the brain.

Are our perceptions always veridical reflections of information in the real world? For the past several hundred years, philosophers and researchers have suspected that the answer to this key question in neuroscience is "no". High-level expectations are believed to play an important role in determining percepts. To a nervous backpacker walking in the woods at night, for instance, every curved stick on the ground might look like a snake.

New experimental findings reported by researchers at the Max Planck Institute for Biological Cybernetics in Tuebingen (Prof. Heinrich Buelthoff and Dr. Isabelle Buelthoff) and the University of Wisconsin-Madison (Prof. Pawan Sinha) in the July issue of Nature Neuroscience confirm some of these suspicions and also extend them in an important way. They reveal that high-level expectations exert such a profound influence on our visual systems that they can actually modify our interpretation of even the basic perceptual attributes that have hitherto been believed to be immune to such influences. These findings have implications for the nature of internal object representations and models of visual processing in the brain.

Interestingly, these findings arose out of an accidental observation. The researchers were attempting to determine whether distorting an object's depth structure while preserving its two-dimensional appearance makes it harder for an observer to recognize the object. The stimuli they were using were biological motion movies of the kind popularized by Johansson in the 1970s. Such sequences are typically generated by having a person in black spandex attach little lights to each of his/her limb joints and then perform simple actions like walking, running or push-ups in a completely darkened room. The resulting sequence shows simply the patterns of motion of the 10-12 points of light (the form of the person is completely obscured by the darkness). Notwithstanding the simplicity of the sequences, observers can immediately recognize the moving lights as constituting a human form. The percept is extremely vivid and therefore seemed to be a good choice for use in the researchers' experiments on recognition.

Three-dimensional data regarding the coordinates of the twelve points over time was manipulated in such a fashion as to completely scramble the depth structure of the underlying human while preserving his two-dimensional appearance from the side. The distorted structure could, for instance, have the knee and elbow points cross each other in depth. This structure was then viewed in 3D with the aid of special stereo-goggles. What the researchers found to their surprise was that not only was recognition completely unaffected by the depth-scrambling but also that these distorted structures did not appear to be distorted. It seemed as if our expectation of what three-dimensional structure a human should have was over-riding the actual depth information being provided by the early stereo processes. Running specially designed experiments with several subjects confirmed these observations. Subjects were significantly more accurate in assessing the depth structure of random dynamic structures (that are presumably not subject to any high-level influences) as compared to a walking human form.

There are two basic implications of these results. The first concerns the nature of the mental representations for dynamic objects. The limited influence of depth information on subjects' recognition responses suggests that the internal representations might emphasize the 2D trace-structure of the object rather than its 3D geometry. Recent reports of 'view-tuned' neurons for 3D objects in the monkey infero-temporal cortex are consistent with this idea. (Coincidentally, these physiological results were obtained by Prof. Nikos Logothetis who has recently joined the Max Planck Institute in Tuebingen.)

The second implication concerns the existence of a top-down influence capable of modulating the information provided by the early depth-perception processes based on binocular disparities. The general idea that top-down influences can affect perception is certainly not a new one. The contribution that this new study makes to this field is in providing evidence that even the very low-level and supposedly purely bottom-up stereo-depth perception is susceptible to top-down influences. These results are especially interesting in the light of some recent computational models of the neo-cortex. These models propose that feedback cortico-cortical projections might provide the channels via which top-down influences might propagate.

Augmenting the conventional feed-forward models of perceptual processes with top-down influences promises to yield a more complete picture of how the brain accomplishes its astounding feats of visual perception and cognition.

###



Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.