M. van den Berg’s research while affiliated with University of Virginia and other places

Rock et al. (Perception, 1992, 21, 779-789) showed that grouping can be computed post-constancy, after discounting a complex scene's variations in illumination. Unpublished studies by Kubovy and his co-workers have shown grouping among heterogeneous dots; hence, constant lightness is not a prerequisite for grouping. These experiments suggest that illuminant complexity can affect the discounting of variations in illumination. Our stimuli were rectangular dot-lattices presented at random orientations in a circular aperture. In these lattices a is the shortest inter-dot distance, which we held constant, and b is the second shortest inter-dot distance, which we varied. We convolved (a) lattices in which the dots and the background were of uniform luminance with (b) sinusoidal luminance gratings, while keeping the luminance ratio of background to dots constant at 1.2. The stripes of the gratings were parallel to the orientation of b. We used a high (3.37 c/d) and low (0.017 c/d) spatial frequency grating (while keeping the maximal change in luminance per pixel constant) to see if observers would disregard the sinusoidal "shadow." The observers were given a phenomenological 4AFC to indicate the grouping in the dot lattice, in three conditions: (i) no grating and (ii) high- and (iii) low-frequency gratings. Discounting did not occur with a high spatial frequency grating (grouping was biased by the orientation of the stripes), but it did occur with a low spatial frequency grating (grouping was not different from grouping in the no-grating condition). This suggests that a complex, non-monotonic, shadow cannot be "discounted," preventing constancy from occurring, whereas a monotonic shadow allows for post-constancy grouping.

In our work on grouping by proximity we noticed effects of lattice orientation on grouping by proximity. In a systematic study of this phenomenon, 25 observers (Os) were presented with 30 repetitions of a hexagonal lattice (in which dots are equidistant along three orientations, 120 degrees apart) at all possible orientations. In a phenomenological 4AFC, Os chose one of four grouping orientations. The 900 trials of each O fell by and large into one of three categories: vertical orientations preferred over all others (V), horizontal orientations preferred over all others (H), and vertical and horizontal orientations preferred over oblique ones (VH). In a more detailed analysis we noted that although some Os' choices were stable, some varied considerably, gradually moving from category V to category H, and back. Our findings are consistent with the operation of two independent mechanisms of roughly equal strength involved in the processing of vertical information: (a) A vertical sensitivity mechanism (VSM), which makes us exquisitely sensitive to deviations from the vertical or to misalignments along the vertical. Increasing the activity of VSM would increase the likelihood V trials. (b) A vertical emphasis mechanism (VEM), which makes vertical lines look longer (and produces the V-H illusion). Increasing the activity of VEM would increase the apparent distance between dots along the vertical and decrease the likelihood of V trials. When VSM is stronger than VEM, V-trials will occur; when VEM is stronger than VSM, H-trials will occur; when VSM and VEM are in equilibrium, small and momentary fluctuations in their strengths could either lead to VH data - or to fluctuations over time.

We tested the perceptual reality of synaesthesia in two phases. During the first phase, four color-grapheme synaesthetes reported on the synaesthetically induced colors of the letters A-Z and the numbers 0–9 in two sessions. We computed consistency within and between the two sessions and identified letters and numbers whose colors are consistently chosen, taking into account the strength of the synaesthetic association and the ability of the computer display to adequately represent the synaesthetic color. In the second phase, we pitted grouping by proximity against grouping by similarity, using letters that induced colors either strongly or weakly. We measured the strength of grouping by proximity and grouping by similarity. Synaesthetes are more likely to group a dot pattern according to the similarity of synaesthetic colors than by proximity. This effect is greater for alphanumeric symbols that induced colors strongly than for alphanumeric symbols that induced them weakly. For control subjects the alphanumeric symbols had no effect on perceptual organization and the difference between strong and weak symbols was absent. The results show that synaesthetes choose induced colors with high consistency and that synaesthetically induced colors affect perceptual organization. Since these results are based on precise quantitative methods, they constitute evidence that synaesthesia is a genuine perceptual phenomenon.

We investigated the spontaneous perception of structure in random dot patterns (RDPs) and the role of the grouping principles of proximity, collinearity, and good-continuation. In three experiments observers reported the perceived organization in RDPs by either circling the groups they perceived or by clicking on the dots that appeared to belong to the same groups. We analyzed these reported organizations were analyzed within-subject and between-subject consistency, and they were used as the foundation for a quantitative model. This model uses graph-theoretic techniques to determine the neighborhoods of dots in the random dot patterns. Within these neighborhoods it computes the strengths of grouping by proximity, collinearity and good-continuation. In a computer simulation the model performed multiple trials with the experimental stimuli and, in a stochastic process, predicted grouping frequencies. Our data were accurately accounted for by a model based on grouping by proximity and good-continuation that computed neighborhoods in a global fashion. In a fourth experiment we tested this model in a new context and investigated the time-course of perceptual organization in RDPs. The predictions of the model captured the observers' performance in this experiment; display duration did not affect the responses. We interpret the results of these experiments and the model-fitting as evidence that the perceptual organization of RDPs is an early, automatic, and global process.