Introduction. Using global shape cues, humans discriminate circles from radial frequency (RF) patterns at hyperacuity levels (Wilkinson, Wilson & Habak, VR1998). Here, we extend our neural model of RF perception (Poirier & Wilson, VSS2005) to account for human perception of symmetry in biologically-relevant shapes (Wilson & Wilkinson, VR2002).

Model. Object position is estimated using large-scale non-Fourier V4-like concentric units, which encode the center of concentric contour segments across orientations. Further processing occurs relative to the estimated object center, providing translation invariance. Shape information is retrieved using curvature mechanisms' responses to visual contours. Curvature mechanisms are scaled with distance from object center, providing scale invariance. Curvature responses were highest at points of maximum curvature, encoding their number, amplitudes, and locations. Symmetry was defined as the correlation of neural curvature responses on either side of a symmetry axis, and the symmetry axis' orientation was defined as the orientation at which symmetry peaked.

Results. Symmetry perception for faces and complex shapes depends on whether curvature extrema positions are symmetrical (Wilson & Wilkinson, VR2002), which depends on the phase-alignment of the component RFs used to create the shape. In our model, symmetry decreased faster with phase-misalignment for stimuli associated with low thresholds in psychophysical experiments.

Discussion. This represents the first model of symmetry shape perception for biologically-relevant shapes (e.g. faces) defined as complex RF patterns. Our model is compatible with recent data on V4 and IT population coding (e.g. Brincat & Connor, NatNeuro2004; Pasupathy & Connor, NatNeuro2002).