Oliver Tošković
Laboratory of Experimental Psychology, Faculty of Philosophy, University of Belgrade |

This symposium is dedicated to topics in the area of perception which were related to the research interests of recently retired professor Dejan Todorović, whose fruitful scientific career marked the work of Laboratory for Experimental Psychology, Faculty of Philosophy, University of Belgrade. Professor Todorović graduated in mathematics in 1975 at the University of Belgrade and received his MA in 1983 and Ph.D. in experimental psychology in 1985 from the University of Connecticut, USA. His postdoctoral positions were in Boston University, USA (1986-1987), Zentrum für interdisziplinäre Forschung, Germany (1994-1995), and Rutgers University, USA (1999). He taught courses on perception, research methods, statistics, and scientific communication, with research interests in experimental studies, mathematical analyses, and computational models of the perception of lightness, space, shape, pictures, and faces. Professor Todorović received third prize for the illusion of the year in La Coruna, Spain (2005), wrote three books, and 95 scientific publications which were cited 2310 times, giving him an H-index of 23 and an i10-index of 34. His creative and inspirational scientific ideas, dedicated teaching and mentoring continues to have a strong influence on perception and experimental research in psychology in Serbia, former Yugoslavia, and beyond.


Oliver Tošković
Laboratory of Experimental Psychology, Faculty of Philosophy, University of Belgrade |

Dejan Todorović

Laboratory of Experimental Psychology, Faculty of Philosophy, University of Belgrade

The distance from an observer to an object on the ground plane (egocentric distance) can be recovered from information about the height of the observer’s vantage point and information about the angular declination of the object (angle subtended by the direction to the horizon and the direction to the object location on the ground plane). Egocentric distances are generally underestimated. It has been suggested that this is due to overestimation of angular declination by a constant factor. With distance held constant, this hypothesis predicts that underestimation will be more pronounced for greater heights of the vantage point. We tested this prediction in two experiments. In Experiment 1 observers judged distances in three conditions: lying on the ground, standing, and standing on a 80 cm high platform. The object whose distance was judged was an experimenter (E1), standing on the ground at 5, 10 and 15 meters from the observer. Observers judged E1’s distance by instructing a second experimenter (E2) to move away from or towards E1 in order to make the E1-E2 distance the same as the E1-observer distance. The E1-E2 direction was perpendicular to the E1-observer direction. We replicated the general underestimation of egocentric distances. However, we found no effect of vantage point height (F=3.1; df=2, 32, p>.05). In Experiment 2, observers were either standing on the ground or on a platform. Their task was to match the distance of a small object at their eye height to the distance of another small object lying on the ground, and vice versa. The positions of the objects were manipulated through a system of cords and pulleys, and their to-be-matched distances were 1, 3 and 5 meters. According to the angular declination hypothesis, distances of objects at eye level should be perceived more accurately, or at least as larger than distances of objects on the ground. A three-way ANOVA revealed all main effect as significant, participant position (F=6.61; df=1, 16, p<.05, η2=.29), object position (F=5.22; df=1, 16, p<.05, η2=.25) and object distance (F=1831.52; df=2, 32, p<.01, η2=.99), but none of the interactions reached significance. Sidak post-hoc tests found no effects of height of vantage point or even an effect contradictory to the angular declination hypothesis. Since, in both of our experiments, we did not find the effect of vantage point height on perceived distance, we were not able to confirm the angular declination hypothesis. 

Keywords: observer’s height, egocentric distance, vantage point, angular declination


Ian M. Thornton

Department of Cognitive Science, University of Malta |

Dejan Todorović has contributed greatly to our understanding of visual illusions. His co-edited volume “The Oxford compendium of visual illusions” (Shapiro & Todorović, 2017) has quickly become a definitive resource in the field and his Perception Lecture article “What are visual illusions” (Todorović, 2020) is a true tour de force. It provides not only a wonderfully insightful review of a very broad range of phenomena, but also a novel and powerful framework for thinking about illusions and a very convincing defence of the position that visual illusions can teach us much about perception in general. In my own work, I have long been interested in various forms of localization errors, where the perceived position of a moving object is shifted in some way from its true position. One puzzle in this field is that laboratory studies of a variety of such phenomena predict much worse real-world performance than we actually see. That is, outside of the laboratory, we are usually able to navigate around the world quite successfully – often at high speed – without collision. Regardless of our own movement, we are also capable of interacting with moving objects, avoiding them or intercepting them, depending on the current task/sport. Could this discrepancy arise because laboratory participants are usually passive observers, but in the real world we are active agents? In this talk, I will focus on one compelling class of illusions known as “motion induced position shifts” (MIPS) and will ask the question of whether introducing action can help attenuate or even overcome this form of illusion. MIPS generally relate to situations in which the global, physical position of a target object is misperceived due to its own local motion. I will present work from a series of studies where participants were given active control over the global position of a target object – steering the object via a joystick or tilt-control – to attempt to accurately position it within a given context. In the crucial comparison, the same observers also made passive judgements, so we were able to directly compare illusion magnitude with and without action. To our surprise, action actually amplified this form of illusion, rather than overcoming it, a finding that has implications for our understanding of vision for perception versus vision for action in the context of illusions.

Keywords: illusions, localization errors, motion induced position shifts, action, perception


Sunčica Zdravković

Laboratory for Experimental Psychology, Department of Psychology, Faculty of Philosophy, University of Novi Sad; Laboratory for Experimental Psychology, Faculty of Philosophy, University of Belgrade |

Visual illusions (Todorović, 2020) are charming perceptual phenomena capturing equally strongly the imagination of laymen and the interest of scientists. A number of compelling illusions can be found in every domain of vision (Shapiro & Todorović, 2016). The kinetic depth effect is an illusory motion effect in which a two-dimensional field of dots appears three-dimensional as soon as it is set into motion. A series of experiments was conducted to define this emerging illusory 3D space. Stimuli were simplified to only 2 dots, which proved sufficient to create the appearance of 3D. The results were compared with predictions from Johansson’s frontal-parallel principle. The principle was confirmed for dots moving on convergent paths, the results were inconclusive for straight paths, and the principle was disconfirmed for divergent paths. Based on the obtained data, we offered an alternative model of illusory space (Zdravković, 2002, 2003). Illusory motion can also appear on completely stationary images. Research will be presented where such images were optimised and used in visual search and cueing paradigms. A “pop-out” effect was found for an illusory motion target placed amongst a variable set size of distractors displayed in a circular array. Control conditions determined that the illusion, rather than structural differences between target and distractors, attracted attention (Thornton & Zdravković, 2020). Similarly, in a modified Posner cueing task paradigm, responses were reliably faster following a valid illusion cue (Zdravković & Thornton, 2017). Both experiments suggest that illusory motion can automatically attract attention. In the above experiments, hue and contrast were essential for creating illusory motion. Hue and contrast themselves can be susceptible to illusory phenomena. In fact, colour and lightness illusions further demonstrate the context effects that lead to illusory percepts (Todorović, 2006), even in the simplest displays (Economou, Zdravković, Gilchrist, 2015). In more complex scenes, the lightness of a target is affected not only by its immediate background but also by non-adjacent portions of the display. We included various grey targets, changed the structure and position of the elements, the curvature of the edges as well as the appearance of transparency. The findings challenged previous explanations based on edge classification and transparency/shadow perception so we offered an alternative model (Todorović, Zdravković, 2014) that will be discussed in the presentation.

Keywords: illusions, kinetic depth effect, illusory motion, lightness, colour 


Dražen Domijan

University of Rijeka, Rijeka, Croatia | 

Dejan Todorović started his theoretical work on lightness perception by studying the Craik-O’Brien-Cornsweet effect (COCE). He devised elegant new variants of COCE, which enabled him to refute many explanations including the non-isomorphic approach, differentiation-integration theory, and cognitive theory. Only the filling-in theory was able to account for Dejan’s variations. These observations led him to develop, in collaboration with Stephen Grossberg, a neural network model of lightness perception. The model consists of two parallel processing systems: a Boundary Contour System (BCS) and a Feature Contour System (FCS). BCS simulates orientation tuning of simple and complex cells, creating a boundary representation of a stimulus. On the other hand, FCS computes luminance ratios by its on-centre off-surround receptive fields and discounts the effect of variable illumination. These systems converge to a common processing stage where diffusive filling-in enables FCS signals to flow across homogeneous surfaces until they are blocked by the output of BCS. The filling-in stage creates a surface representation that is hypothesised to be isomorphic to subjective perception. Grossberg and Todorović tested the proposed model through a comprehensive set of computer simulations on 1-D and 2-D input images. They showed that the model is capable of explaining many important phenomena including lightness constancy, lightness contrast and assimilation, the Hermann grid, COCE, and many others. Their neural network is a cornerstone in modelling lightness perception, which inspired many later theoretical developments including my own modelling attempts. In subsequent work, Dejan focused his attention on the role of junctions in modulating lightness perception. He designed a new illusion, which now bears his name, to show that perceived lightness departs in an opposite direction from what would be predicted by simple low-level mechanisms such as lateral inhibition. He suggested that the illusion arises from depth segregation of white and black surfaces induced by T-junctions. Todorović’s illusion now serves as a critical test for any serious computational model of lightness perception. To conclude, I reviewed just a small part of Dejan’s diverse work, but I hope to illustrate how profound and creative his contributions to the study of visual perception and visual illusions have been.

Keywords: lightness perception, lightness illusions, neural network model, computer simulations 


Slobodan Marković

Laboratory of Experimental Psychology, Faculty of Philosophy, University of Belgrade |

Dejan Todorović

Laboratory of Experimental Psychology, Faculty of Philosophy, University of Belgrade

The classical Metelli’s patterns can be classified in the three categories. The first category encompasses phenomenally transparent patterns: observer sees the transparent figure (e.g., rectangle) centered in front of opaque bicolor ground. The second category includes phenomenally shadowed patterns: observer sees bicolor surface with rectangular shadow on the center. The third category includes mosaic patterns: observer sees mosaic composed of adjacent surfaces with different gray-levels. Our demonstrations show that the similar configurations of surfaces don’t produce an equally clear transparency effect. In the case when the central rectangular area of pattern satisfies the “transparent conditions”, it will be clearly seen as a transparent frontal figure. However, when the same conditions are satisfied in the peripheral frame-like area, the transparency will not be seen so convincingly. Thus, the figural factors, i.e., the distributions of surfaces with different gray-levels, determine the strength of phenomenal transparency: the smaller central part of the visual field figure is more likely than the peripheral frame seen as the figure. In the present study the figural factors are controlled by introducing the model of two partially overlapping squares. According to this model both squares have equal probability to be seen as a figure. In our experiment, participants viewed two partially overlapping squares, one lighter and the other darker, with the luminance of the overlapping portion being varied so that it was more similar to the lighter or darker square, or had a luminance between the brightness of the two squares. Participants were asked to determine which of the two partially overlapping squares is transparent, i.e., which is perceived as closer to the observer, and which is opaque, i.e., located in the background. The results showed that, although in all three conditions the surface ratios satisfy Metelli’s rules, the main factor that influenced the responses of the participants was the similarity of luminance: a square is perceived as transparent when the luminance of the overlapping part is more similar to its surface, while in the condition of intermediate luminance responses of the participants are bistable. These results do not contradict Metelli’s rules that determine whether transparency will be seen or not, but they specify the conditions that determine what will be seen as transparent and what will be opaque.

Keywords: transparency, mosaic, figure, ground, gray, overlapping