Optical
illusions are artifacts of sensory processing.
Though misperceptions, they highlight principles and objectives that
govern normal neural processing. Therefore, a comprehensive theory of vision
should be able to account for optical illusions. Here I will tell about retinal
mechanisms hacked by such types of stimuli.
I strongly
recommend arming yourself with a plan of the retina from a Figure 1.
The retina
has to provide the rest of the body with sufficient information about the
visual environment. On the other hand, we would like it to consume as little
energy as possible. To achieve both of these goals, the retina has to be
efficient and transmit only important information.
The visual
environment is an aggregate of figures and backgrounds. We are interested in
their separation i.e. evaluating differences between them. A way to perform
this task efficiently is to transmit information about contrast instead of an
absolute light intensity. Contrast characterizes how objects differ in a
reflecting light. This is an internal property of an object, thus it allows
discriminating objects regardless of the absolute intensity of the incident
light. Hence, there is no need to send information about it and this optimizes
retinal performance.
How to
estimate contrast? Figures are surrounded by other figures and background, so
we estimate contrast by subtracting the signal of a given photoreceptor from
the signals collected by its neighbors. This process is mediated by horizontal
and amacrine cells and called “lateral inhibition”. Consequentially, bipolar
cells, which collect signals from photoreceptors, and ganglion cells, which
transmit it further to the brain, exhibit antagonistic center-surround
receptive field arrangements. That is, these cells are excited by stimuli in
their centers, and inhibited by ones in their surround. So essentially they
transmit difference between center and surround.
The illusion
shown in Figure 2 exploits this processing strategy. Although painted uniformly
throughout, the rectangle shown appears lighter on the left and darker on the
right. As we know now, the illusion occurs because instead of perceiving an
objects absolute brightness we perceive how it differs from its surroundings.
As the background is dimmer on the left and lighter on the right, the contrast
between background and rectangle is higher leftwards. Thus, object appears
brighter.
Another
illusion based on this principle - perceiving not intensity, but contrast, is
Hermann grid. In Figure 3, with a quick glance we can note grey blobs on the
intersections of the grid.
There is a
sudden increase in background luminance when we shift our gaze to an
interception.
So
excitation of the receptive field center remains the same, whereas inhibitory
signals from the background suddenly become stronger, thereby darkening the
perceived shade of the interception.
It is notable that is even simplest
retina-inspired neuromorphic devices (Mahowland and Mead 1991) are susceptible
to Herman grid illusion. It highlights that illusory effects arise from the
very basics, almost nuts and bolts, of retinal processing.
Light Adaptation
Although
lateral inhibition removes information about absolute light intensities from
the photoreceptor outputs, they still need to deal with huge variations in the
level of their inputs. The reason for this is very simple: light intensity can
vary over 7 orders of magnitude, but the response range of photoreceptors is
quite limited. To overcome this issue photoreceptors decrease their sensitivity
upon stimulation. This process underlies afterimage illusions.
Let’s consider the image in Figure 4. If you
look at it for 30 seconds to 1 minute and then direct your gaze to a white
object, you will perceive cyan inscription over a magenta background. This is
called an “afterimage”. It is a retinal effect, so if you will move your eyes
along a white object, the cyan inscription will move as well. Humans possess
three types of photoreceptors for color vision: red, green and blue. When you stare
at a red inscription, red-sensitive photoreceptors gradually decrease their
sensitivity. That’s why white objects stimulate them to a lesser extent than it
does with green and blue-sensitive photoreceptors. We perceive color as a
relative activation of opponent channels: red versus green, yellow versus blue,
and white versus black. So decreased activation of red channel shifts the
overall color picture to its complementary color, that is - cyan. The magenta
color of the background can be explained by similar logic.
Dress & Color Constancy
Back in a
day, the dress in figure 5 caused a lot of dispute about its color on the
internet.
Despite the
true color of the dress being blue and black, roughly 40% of people perceive it
as white and gold or blue and gold (Lafer-Sousa et al. 2015). This photo
highlights stunning and bizarre individual differences in color perception.
Spectral
composition of a light reflected by an object, also known as color, is one of
its internal properties. Thus, it can help with object identification. Hence,
we would like to minimize the influence of illumination on its perception.
Indeed, even though evening illumination is quite red compared to midday light,
the perceived colors of your car or of tree leaves remain unchanged. This
phenomenon is called color constancy, and it is this mechanism that is confused
by the dress.
To have the
colors of an object constant under various illuminations we simply need to
discard illumination. According to experimental data, that is where differences
in the dress color perception arises. First of all (Aston and Hurlbert 2017),
people who see the dress as blue and black have a tendency to estimate
illumination as yellowish, and those who perceive the dress as white and gold
think that illumination is blue. Moreover, when the cues to the illumination
are enhanced (Lafer-Souza et al. 2015) ambiguity in a dress perception
disappears.
Although color constancy engages various
cortical mechanisms, essentially it is a figure-background segregation task.
Hence, it roots lay in a lateral inhibition, which initiates within the retina
and is repeated at the various levels in the visual cortex.
Perceiving the Present
In contrast
to our eukaryotic brothers, we animals are able to move and movement vastly
increases the pace of life. This is why the nervous system emerged: to guide,
coordinate and navigate us. As Pavlov
brilliantly demonstrated, the nervous system anticipates rather than reacts. In
an ever-changing world, being ready for today means being late because
preparation takes time.
Vision is
there to create a map of an environment. Signal processing and transmission
take time. That leads to a delay of roughly 100ms between stimulus onset and
elicited perception.
This is
mainly due to the slow transduction processes of photoreceptors that transduce
photons into electrical signals. The visual system needs to compensate for this
delay somehow if we really want to perceive the present.
According to
Changizi et al. (Changizi et al. 2008) confusion of this mechanism leads to a
vast number of optical illusions, like the classical geometrical illusions.
A
combination of these illusions (Herring, Orbison and Ponzo) are shown in Figure
6. Motion leads to a displacement of the objects. In this case, compensation
for the delay is critically important. To compensate, we need to estimate how
the scene will look in the following 100 ms. To do so we use some cues to
figure out the direction and speed of movement and then to model how it will
affect an object.
The optical flow induced by motion engenders
radial “smear’’, which is mimicked by the radial display shown in Figure
6. It makes us think we move forward.
What we see is not an actual picture, but how it will look like in the next
moment. When you pass through the door, its opening goes sideways. That’s why
the vertical lines in Herring illusion appears non-parallel. Surprisingly, we
become closer to objects in a direction of motion while moving. As a result,
the retinal area on which their image is projected, also known as angular size
increases. This causes distortion of the squares shape in Figure 6, a phenomena
referred to as an Orbison illusion. Same
reason underpins Ponzo illusion as well.
Figure
6 M. A.
Changizi et al./Cognitive Science 32 (2008)
This
activity moves at the true location of the object or even along its leading
edge (Berry et al. 1999). It means that the delay is already compensated for on
the level of retinal output. The underlying mechanism is as follows: Ganglion
cell activation reports that an object is passing by. Earlier we discussed
receptive fields - areas from which neurons collect their inputs. Receptive
fields are extended in space. Therefore, a moving object activates some of the
ganglion cells ahead of its motion. It activates ganglion cells when just
crosses the edge of the receptive field, but not completely passed by yet. So in a way, the spatial extension of the
ganglion cell receptive fields compensates for the delay. One might note that such spatial extension
alone is not enough for a proper delay cancelation. Since it also implies
activity of ganglion cells when object have already passed by. This issue is
overcome by using transient cells as they only activate briefly when an object
crosses their receptive field, then quickly return to their non-activated
state.
Intuitively
and based on a personal experience with tracking objects, we might suggest that
the quality of extrapolation and delay cancelation should depend upon a
contrast, which is indeed the case. While interesting, this is not a very surprising
result. We already know that due to
lateral inhibition, the signal fed into ganglion cells is simply contrast. Weak
signals i.e. weak contrast will just fail to activate ganglion cell in a proper
way regardless of the speed, direction or any other parameter of motion.
There are
more than 20 different types of ganglion cells. Each type senses its particular feature. However, regardless of those
features all ganglion cells inevitably sense contrast and this sometimes
creates a mess. For instance, let’s consider the next example, the “Plaid
stimulus”. This stimulus consists of two gratings passing through each other.
When asked about the direction of plaid motion we Homo sapiens show a bias
towards grating with the higher contrast as this contrast causes stronger
activation of the corresponding direction-selective cells. This subset of
ganglion cells only activate when an object moves in a certain direction along
vertical and horizontal axes.
There is
also a certain type of retinal ganglion cells, which detect an object
approaching.
It is
contrast-sensitive as well. I suppose that’s why the left part of a rectangle
in Figure 7 appears slightly closer.
Indeed, higher contrast causes stronger activation of approach-sensitive
cells in the part of the retina where an image of this object is projected. So
in terms of ganglion cell firing rates, it looks like the stimulus on the left
is approaching us.
As you
remember, to perceive the present the visual system has to model how a scene
will change in the next moment. And, as you might also note, objects which
we approach usually becomes closer. In addition, its angular size increases. That’s
why the circle with the higher contrast in Figure 7 looks bigger.
Figure
7 M. A.
Changizi et al./Cognitive Science 32 (2008)
Conclusion
Here I discussed some neuronal processes behind
optical illusions. Namely light adaptation, lateral inhibition and compensation
of neuronal delays. These are all very
general, basic and subliminal mechanism of visual processing. The bottom line
here is that optical illusions stem from the organization of the nervous system
itself, from its ultimate limitations and concerns. These incorrect-perceptions
hack-strategies in most cases are consequences of optimizing neuronal
performance. So perception of illusions is a downside of neuronal efficiency,
which meanwhile nicely illustrates this efficiency.
References
Aston, S.,
& Hurlbert, A. (2017). What #theDress reveals about the role of
illumination priors in color perception and color constancy. Journal of Vision,
17(9):4, 1–18, doi:10.1167/17.9.4.
Berry, M.J.,
2nd, Brivanlou, I.H., Jordan, T.A., and Meister, M. (1999). Anticipation of
moving stimuli by the retina. Nature 398, 334–338.
Changizi
M.A., Hsieh A., Nijhawan R., Kanai R., Shimojo S. Perceiving the Present and a
Systematization of Illusions. Cognitive Science 32 (2008) 459–503
Lafer-Sousa,
R., Hermann, K. L., & Conway, B. R. (2015). Striking individual differences
in color perception uncovered by ‘‘the dress’’ photograph. Current Biology,
25(13), R545–R546.
Mahowald
Misha, Mead Carver “The Silicon Retina” Scientific American, May 1991
