Introduction: The Problem of Subjective Contours
Subjective contours are illusory edges perceived in the absence of physical luminance changes, famously exemplified by the Kanizsa Triangle. In 1976, Shimon Ullman proposed that the human visual system completes these shapes using two circular arcs that meet smoothly. His model relied heavily on "extensibility"—the geometric principle that if a curve is completed from a point further along its path, the resulting shape should remain identical.
However, Brady, Grimson, and Langridge argue that extensibility is a "red herring." Drawing parallels to mathematical typography (such as Donald Knuth’s Metafont), they point out that enforcing extensibility often conflicts with other essential properties like smoothness and isotropy.
Methodology: Minimizing Quadratic Variation
The authors propose a shift from pure geometry to control theory. They hypothesize that subjective contours are a byproduct of the brain's 3D surface interpolation machinery. Specifically, they suggest the visual system minimizes a performance index related to the curve's energy, specifically quadratic variation ($f_{xx}^2$).
Using a Hamiltonian approach to minimize this variation, they derive a cubic spline solution for boundary angles $\phi$ and $\theta$ over a distance $t$: $$y = \frac{x^3}{t^2}(\tan \theta - \tan \phi) + \frac{x^2}{t}(\tan \phi - 2 \tan \theta) + x \tan \theta$$ When the boundary angles are equal ($\phi = \theta$), the cubic term vanishes, and the solution simplifies to a parabola.
Key Results and the Indistinguishability Problem
A significant contribution of this work is the demonstration that for typical angular extents, the cubic spline model and Ullman’s two-circle model produce nearly identical results.
This "indistinguishability" explains why previous psychophysical experiments failed to favor one model over the other. By linking 2D contour completion to 3D surface interpolation (a known task of the visual system), the authors provide a more integrated biological theory.
Limitations and Future Outlook
While the cubic model is mathematically elegant and biologically plausible, the authors acknowledge that resolving the debate through experimentation is difficult due to the similarity of the curves. They suggest that testing extreme angular cases might provide the necessary differentiation. Furthermore, while this paper focuses on shape, the authors hint that the perceived intensity of illusory surfaces is likely governed by the same underlying interpolation mechanisms.
Summary updated