Tilt Aftereffects in a Self-Organizing Model of the Primary Visual Cortex

James A. Bednar and Risto Miikkulainen

Visual illusions and aftereffects have long been studied by psychologists and vision researchers. Seeing how your brain can be fooled is often entertaining, and even scientists have been known to enjoy what is commonly called "fun". More importantly, these apparent failures of the brain can offer insight into how visual processing is carried out, which can be fun in its own brainy way.

You can reproduce one interesting type of visual aftereffect, the tilt aftereffect, using your very own eyes as follows.


TAE Demo


Fixate your gaze upon the circle inside the central diagram above for at least thirty seconds, moving your eye slightly inside the circle to avoid developing strong afterimages. Then quickly fixate upon the diagram at the left. The vertical lines should appear slightly tilted clockwise; this phenomenon is called the direct tilt aftereffect. If you instead fixate upon the horizontal lines at the right, they should appear barely tilted counterclockwise, which is called the indirect tilt aftereffect. The indirect effect is small, and is not always noticeable outside the laboratory. On the other hand, if the direct TAE didn't work, try again for a longer time. If it still doesn't work, you probably have some horrible brain disorder, which you should see a doctor about, or a problem with your video display, which you should see a computer professional about. A version of this page with smaller pictures is also available.

The above demonstration shows what happens at two different orientations 90 degrees apart. When psychologists measure the effect in the laboratory, they find that the perceptual errors vary systematically with the angle between the adapting and test orientations:

TAE w.r.t. Angle

(Data is from Mitchell and Muir, Vision Research 16:609-613, 1976). Just as in the demonstration above, there is no effect on the perceived orientation of a figure while you adapt to it (it is zero at 0 degrees), but nearby orientations are strongly shifted away from the adaptation orientation (the direct effect, shown by the peaks at 5-10 degrees). Distant orientations (40 to 85 degrees) show a more-variable but generally opposite indirect effect.

Most orientation illusions and aftereffects like these are thought to arise in the the primary visual cortex. The primary visual cortex is an area a little larger than a quarter at the back of your head, and it is (despite its location!) the first region of the cortex that processes visual information. It is also the first region the encodes visual input in terms of orientation. Thus these particular illusions serve as test cases for theories about how the primary visual cortex works.

Several researchers have proposed that the aftereffects result from strengthening of lateral inhibitory connections between cortical neurons receiving visual input. That is, the neurons processing a particular range of orientations might begin to suppress (inhibit) each other more strongly during prolonged viewing, which could help to highlight any changes in orientation that occur later. As a side-effect, the stronger inhibitory connections would also cause the overall estimated orientation of later patterns to change, which is experienced as an aftereffect.

Recent large-scale computer models of brain function allow this general hypothesis to be tested in much greater detail, in order to figure out exactly what the brain is doing and why it is doing it. Our results from such a model are available in a series of animations which show how the brain could be adapting to the visual input over short time scales, and how that adaptation could affect subsequent visual processing. These results provide detailed support for the lateral inhibition theory of direct tilt aftereffects, as well as providing a simple and novel explanation for indirect effects.

The model we used, RF-LISSOM (Sirosh and Miikkulainen, 1994), is a self-organizing model of laterally connected orientation maps in primary visual cortex. This model is the only one currently available that can suitably model the tilt aftereffect, because others lack specific lateral connections or do not self-organize into realistic primary visual cortex structures. The model was originally designed to account for an infant's visual cortex development, but here we show that the same processes that drive development can also account for tilt aftereffects in the adult. These results provide a unified computational explanation of cortical self-organization and both direct and indirect tilt aftereffects in the primary visual cortex.

Neural Computation cover
After checking out the animations, more details about the aftereffect research can be found in our 2000 Neural Computation paper, available online as PostScript and HTML. A preliminary summary was presented at the 1997 Cognitive Science Conference at Stanford and published in the proceedings, but that paper has now been superseded by the Neural Computation paper. Much more background information, including an in-depth survey of related and future work, is available as a 1997 Master's thesis, which is online in HTML. If you want general background on the RF-LISSOM model, two good starting points are a 1996 HTML book article (easy to read online) and a 1997 book chapter (monochrome PostScript; easy to print out). The book chapter also contains some very preliminary aftereffect results, but these were long ago superseded by conference and journal papers.

Keywords: tilt aftereffect, tilt after-effect, tilt aftereffects, tilt after-effects, TAE, tilt illusions, TI, orientation aftereffects, orientation after-effects, orientation illusions, visual illusions, aftereffects, illusions, primary visual cortex, self-organization, orientation maps, neural networks, perception, vision, visual development, visual function, computational neuroscience.

tae-demoTilt Aftereffect Demo (35KB)