A new class of illusion underscores the highly constructive nature of visual perception


A new class of illusion, developed by a visual artist and a psychology researcher, underscores the highly constructive nature of visual perception.

The illusion, which the creators label “Scintillating Starburst,” evokes illusory rays that seem to shimmer or scintillate—like a starburst. Composed of several concentric star polygons, the images prompt viewers to see bright fleeting rays emanating from the center that are not actually there.

“The research illustrates how the brain ‘connects the dots’ to create a subjective reality in what we see, highlighting the constructive nature of perception,” explains Pascal Wallisch, a clinical associate professor in New York University’s Department of Psychology and Center for Data Science and senior author of the paper, which appears in the journal i-Perception.

“Studying illusions can be helpful in understanding visual processing because they allow us to distinguish the mere sensation of physical object properties from the perceptual experience,” adds first author Michael Karlovich, founder and CEO of Recursia Studios, a multidisciplinary art and fashion production company.

The authors acknowledge that the visual effects of this illusion are superficially similar to a number of previously described effects of other, grid-based illusions. However, their Scintillating Starburst, unlike known visual illusions, evokes a number of newly discovered effects, among them that fleeting illusory lines diagonally connect the intersection points of the star polygons.

This stimulus is made up of several concentric pairs of scaled star polygons. Most observers perceive fleeting rays, beams, or lines emanating from the center that appear to be brighter than the background. Credit: Michael Karlovich, Recursia LLC

To better understand how we process this class of illusion, the researchers ran a series of experiments with more than 100 participants, who viewed 162 different versions of the Scintillating Starburst, which varied in shape, complexity, and brightness.

The research participants were then asked a series of questions about what they saw—for instance, “I do not see any bright lines, rays, or beams,” “I maybe see bright lines, rays, or beams, but they are barely noticeable,” and “I see bright lines, rays, or beams, but they are subtle and weak.”

The authors found that the confluence of several factors, including contrast, line width, and number of vertices, matters.

“In particular, a large number of prominent intersection points leads to stronger and more vivid rays, as there are more cues to indicate the implied lines,” observes Wallisch.

Thus, this research illustrates how the brain “connects the dots” to create one’s subjective reality, even on the perceptual level, highlighting the constructive nature of perception.

Illusions have played a key role in understanding the principles of perceptual processing (Purkinje, 1825; van Buren & Scholl, 2018), although this notion is not entirely without detractors (Braddick, 2018; Rogers, 2019). One important reason why illusions can be helpful is that they allow us to distinguish the mere sensation of physical object properties from the perceptual experience. This is perhaps clearest in the case of illusory contours.

If contours are defined by a sharp change in luminance, it is hard to tell whether the observer “directly” perceives objects in the environment “as they are” or if the percept is constructed in the mind of the observer, because the output of a photometer and perceptual judgments agree (Gibson, 1978; Fodor & Pylyshyn, 1981). Conversely, illusory contours are not defined by changes in luminance, and would be invisible to a photometer, whereas human observers readily perceive them. One example of this is Kanisza’s triangle, which most observers perceive as a bright triangle occluding three black circles – as opposed to the three black pac-man-like shapes are actually defined by luminance (Kanisza, 1955).

This phenomenon is an instantiation of a broader principle, by which organisms interpret scenes in ways that require the fewest assumptions, or are probabilistically most plausible (Koffka, 1935; Maniatis, 2008). In this case, it is more likely that 4 common shapes (circles and triangles) are in this scene than 3 uncommon shapes (pac men) that are arranged in this fashion by sheer coincidence.

There are even neurons in prestriate cortex that seem to take such contextual considerations into account and respond to illusory contours as if they were luminance defined contours (von der Heydt et al, 1984).
Here, we introduce a related, but novel type of stimulus that evokes illusory rays that seem to shimmer or scintillate, see figure 1.

Figure 1. The “scintillating starburst” stimulus. This stimulus is made up of several interlocking polygons with faces that bisect each other (see Method section for details). Most observers perceive fleeting rays, beams or lines emanating from the center that appear to be brighter than the background.

To our knowledge, this class of ray patterns has not been previously described (Bach, 2005; Shapiro & Todorov, 2017). The purpose of this manuscript is to introduce a novel type of ray pattern and to explore the space of stimulus parameters that modulate subjective experiences of ray strength in one instantiation of this class – the “scintillating starburst” in the hopes of understanding what neural and cognitive processes might underlie this phenomenon.


The “scintillating starburst” stimulus seen in figure 1 elicits transient shimmering rays that seem to be emanating from the center for most observers. This stimulus was designed in the following way. We used MATLAB (Mathworks, Inc., Sherborn, MA) to construct a regular heptagon (central angle = 51.143 degrees, rounded to 3 decimals), assuming a line width of 2 points, see figure 2A. Figure 2B was created by adding a 2nd, identical heptagon rotated by 25.71 degrees (pi /7 radians) so that the faces of the two heptagons bisect each other.

We call the image element depicted in Figure 2B a “strand”. Figure 2C was created by adding a scaled version (89.95%) of the heptagon in Figure 2A to itself. Figure 2D was created by the same principle, but now, the 2nd strand is again rotated such that the vertices of the heptagon faces that make up this strand also bisect each other. The scale factor was chosen so that the two strands just touch each other, in other words the vertices of the smaller heptagon bisect the midpoints of the larger heptagon.

We call the image element depicted in figure 2C and D an “inducer ring” (not-bisecting and bisecting versions, respectively). The image in figure 2E was created by adding another not-bisecting inducer ring to 2C such that the 2nd ring is a scaled down version (66%).

The image in 2F was similarly created by adding another bisecting inducer ring to 2D, using the same scale factor. Thus, 2E and 2F feature two not-bisecting and bisecting inducer rings each, respectively. Finally, figures 2G and 2H were created by adding an additional inducer ring using the same scale factor, so that these images are featuring 3 inducer rings. One could repeat this principle to create stimuli with even more inducer rings. One could also vary the number of vertices in the polygon.

Figure 2. An illustration of how the scintillating starburst stimulus was constructed. Top row: Non-bisecting stimuli, Bottom row: Bisecting stimuli. A: A single heptagon. B: Two heptagons with faces that bisect each other, which we call a “strand”. C: Two scaled heptagons – a “non-bisecting” strand. D: Two scaled strands of bisecting heptagons, making up one “inducer ring”. E: Two scaled non-bisecting rings. F: Two scaled bisecting rings. G: Three scaled non-bisecting rings. H: Three scaled bisecting rings. This stimulus corresponds to the “scintillating starburst stimulus” shown in figure 1, but on a white background and smaller.

In fact, we did just that to parametrically create visual stimuli varying in 5 stimulus dimensions: 1) Number of vertices of the polygon [3 levels: 3 5 7], 2) Contrast [3 levels:
0.1 0.55 1], 3) Line width of the innermost ring [3 levels: 0.5 1 1.5 points], 4) Number of inducer rings [3 levels: 2 4 6], 5) Whether the strands of the inducer rings bisect each other [2 levels: Yes and no].
Fully crossing these stimulus dimensions yields the set of 162 unique stimuli we used in this study. All other stimulus dimensions were kept the same across all stimuli so as to not further increase the number of stimuli in the set. All stimuli were presented on a white background at a size of 600×600 pixels, delivered remotely.
See figure 3 for an illustration of a representative sampling of this space as well as the average experienced ray strength (RS) evoked by these stimuli in our sample of observers. As you can see, our set of stimuli evoked a wide range of responses andwere effective in modulating the experience of the observers.

The reason we picked these stimulus dimensions and not others is to avoid the curse of dimensionality – there is a vast space of visual parameters to vary, many of which have no impact on the effect. For instance, extensive piloting suggested that color, size or uncoiling the rings to yield non-radial ray patterns did not seem to affect experienced ray strength much.

Figure 3. A representative sample of 20 out of the 162 stimuli used in this study. Pictograms of the stimuli are arranged in increasing order of average ray strength reported by our sample of observers. Top rows: Weak ray strengths. These rows are predominantly made up of triangular polygons, low contrast stimuli and non-bisecting stimuli. 2nd to last row: Stimuli with moderate experienced ray strength – predominantly bisecting pentagons. Bottom row: These 5 stimuli evoked the strongest ray experiences. These stimuli consist of high contrast bisecting heptagons with several rings. This stimulus is made up of several interlocking polygons that bisect each others faces (see Method section for details). Most observers perceive fleeting rays, beams or lines emanating from the center that appear to be brighter than the background. The label above each icon indicates the level of the independent variable, in order (number of vertices per polygon, contrast, line width, number of rings and whether the polygon faces bisect each other. The RS value below each icon denotes the average experienced ray strength.

More information: Michael W. Karlovich et al, Scintillating Starbursts: Concentric Star Polygons Induce Illusory Ray Patterns, i-Perception (2021). DOI: 10.1177/20416695211018720


Please enter your comment!
Please enter your name here

Questo sito usa Akismet per ridurre lo spam. Scopri come i tuoi dati vengono elaborati.