Gregory Francis¹ (), Mauro Manassi²; ¹Purdue University, ²University of Aberdeen

In serial dependence, our percepts (such as orientation) are biased towards previously presented stimuli. Serial dependence was proposed as a mechanism to facilitate perceptual stability and improve efficiency by compensating for variability in visual input. Despite a large amount of research on the topic, the field lacks specific underlying mechanisms to explain how this attractive bias occurs. Here, we show that serial dependence in orientation perception is naturally predicted with a biologically plausible model of visual perception that identifies Gabor stimulus orientation by observing the interplay between perceptual filling-in of surfaces and grouping of orientation-sensitive contours. At its core, the model hypothesizes a serial process that searches for the grouping orientation that maximizes the difference between dark and light filled-in surfaces. A reasonable starting point for this search process is the orientation in recent visual experience. With this assumption, we show model simulations that account for many basic characteristics of serial dependence. Serial dependence gradually decays over time for trials further in the past (up to 3 trials back; temporal tuning). Serial dependence strength peaks at around 15 degrees of relative orientation difference between the current and previous trials and it decreases as the relative difference differs (feature tuning). Consistent with recent proposals of serial dependence as an optimizing strategy, the model also predicts that orientation discrimination performance improves when serial dependence occurs. The model can also successfully explain attentional tuning and spatial tuning properties as being due to changes in the probability of using the grouping orientation from the previous trial as the start of the search process for the current trial. Together, our model provides extensive evidence for a biologically plausible grouping mechanism that explains the basic characteristics of serial dependence effects.