Topological perception: Holes in an experiment

John M. Rubin, Nancy KanwisherPerception 1985 · DOI

Introduction

In 1982, researcher Lin Chen proposed a provocative theory: the human visual system is hard-wired to prioritize topological features—such as holes or connectedness—over basic geometric properties like shape. Chen’s evidence relied on "same/different" discrimination tasks where subjects found it significantly easier to distinguish a disk from an annulus (a ring) than a disk from a square. This suggested that topology is a "primitive" visual feature processed before specific geometry.

The Problem: The Luminous Flux Confound

Rubin and Kanwisher (1985) hypothesized that Chen’s results were driven by a confounding variable: luminous flux. In Chen's original experiments, the "topologically distinct" shapes often had significantly different total black areas compared to the "geometrically distinct" controls. Because the primate visual system is highly sensitive to changes in total light intensity, the perceived difference might simply be a reaction to brightness rather than mathematical topology.

Methodology

To test this, the authors replicated Chen’s experiment using a tachistoscope to flash images for brief 5-millisecond intervals. They designed four stimulus pairs to decouple topological differences from differences in black area (luminous flux):

  • Pair A: A disk-with-two-holes vs. an annulus (Topologically different; zero flux difference).
  • Pair B: An annulus vs. a horseshoe (Topologically different; small flux difference).
  • Pair C: A disk vs. a square (Topologically identical; moderate flux difference).
  • Pair D: A disk vs. an annulus (Topologically different; large flux difference—Chen's original stimulus).

Results and Analysis

The results demonstrated that discriminability was almost entirely dependent on the difference in black area, not topology.

Table 1: Stimulus pairs and corresponding experimental results showing the correlation between black area difference and discrimination accuracy.

As shown in the data, subjects struggled to distinguish Pair A (33.6% accuracy) despite its topological difference, because the flux was balanced. Conversely, Pair C—which had no topological difference—was easier to distinguish (44.4% accuracy) than either Pair A or B because it had a larger flux difference. Pair D, which combined both topological and large flux differences, yielded the highest accuracy (63.5%).

Conclusion

The study suggests that "luminous flux" is the true primitive signal in these rapid discrimination tasks. When area is controlled, the alleged advantage of topological perception disappears. While topology remains a significant concept in higher-level vision, it does not appear to be the "primitive" foundation of the human visual system as previously claimed. The findings advocate for parsimony in psychological models, favoring basic light detection over complex mathematical processing in early vision.

Summary updated

BibTeX
@inproceedings{rubin1985topological,
  author = {John M. Rubin and Nancy Kanwisher},
  title = {{Topological perception: Holes in an experiment}},
  booktitle = {Perception},
  year = {1985},
  doi = {10.3758/bf03202856},
}
Figure