Abstract
Convexity preference is one of the factors that influence figure-ground organization. However, in certain conditions, the convexity preference can be suppressed and non-convex regions might be perceived as figural. It has been suggested that consistency of surface properties plays a key role for this reversal. For example, if a convex region is in the middle of another surface and has the same color/texture as the background, it is often perceived as a hole. The preference of convex regions in repetitive columnar configurations is also reduced if the concave regions have inconsistent colors (Peterson & Salvagio, 2008, Journal of Vision, 8(16), 4.1–13). Importantly, Zhou et al. (2000, Journal of Neuroscience, 20(17), 6594–6611) showed that many border-ownership (BOWN) sensitive neurons in V2/V4 were also sensitive to contrast polarity. Accordingly, Zhaoping (2005, Neuron, 47(1), 143–153) developed a model in which BOWN signals are enhanced when they are consistent in both the ownership and the contrast polarity. Inspired by her model, we first developed a simplified algorithm to compute BOWN that exhibit the convexity preference. It successfully reproduced illusory contour perception (DISC model, Kogo et al, 2010, Psychological Review., 117(2), 406–439). We, then, tested the performance of the model which also reflects the consistency of the surface colors at the location of the signals as in Zhaoping's model. We report that this approach gives extremely robust responses to various images with complexities both in shapes and in depth orders such as the examples mentioned above, suggesting the importance of this approach for BOWN computation. We further investigated, 1: the role of contrast insensitive BOWN signals, 2: the role of concavity preference algorithm, and 3: the effect of inhibitory connections. We will report how these factors affect the model's responses to reproduce figure-ground perception of complex figures.
Meeting abstract presented at VSS 2014