Neural dynamics of feedforward and feedback processing in figure-ground segregation

Oliver W. Layton, Ennio Mingolla, Arash YazdanbakhshFrontiers in Psychology 2014 · DOI

Abstract

Determining whether a region belongs to the interior or exterior of a shape (figure-ground segregation) is a core competency of the primate brain, yet the underlying mechanisms are not well understood. Many models assume that figure-ground segregation occurs by assembling progressively more complex representations through feedforward connections, with feedback playing only a modulatory role. We present a dynamical model of figure-ground segregation in the primate ventral stream wherein feedback plays a crucial role in disambiguating a figure's interior and exterior. We introduce a processing strategy whereby jitter in RF center locations and variation in RF sizes is exploited to enhance and suppress neural activity inside and outside of figures, respectively. Feedforward projections emanate from units that model cells in V4 known to respond to the curvature of boundary contours (curved contour cells), and feedback projections from units predicted to exist in IT that strategically group neurons with different RF sizes and RF center locations (teardrop cells). Neurons (convex cells) that preferentially respond when centered on a figure dynamically balance feedforward (bottom-up) information and feedback from higher visual areas. The activation is enhanced when an interior portion of a figure is in the RF via feedback from units that detect closure in the boundary contours of a figure. Our model produces maximal activity along the medial axis of well-known figures with and without concavities, and inside algorithmically generated shapes. Our results suggest that the dynamic balancing of feedforward signals with the specific feedback mechanisms proposed by the model is crucial for figure-ground segregation.

Introduction: The Challenge of Figure-Ground Segregation

Separating an object (figure) from its background (ground) seems effortless, but it is computationally complex. While detecting edges is straightforward, "filling in" the interior of a shape—especially one with complex concavities like a 'C' shape—is much harder. Neurophysiological data shows a "late component" in brain activity: neurons in early visual areas (like V1) show enhanced firing for a figure's interior only after a delay of 80–100ms. This suggests that higher-level feedback is essential to disambiguate the interior from the exterior.

The Teardrop Model: A Multi-Stage Dynamical Approach

The authors developed a multi-stage dynamical model representing the primate ventral stream, incorporating units that model cells in V4 and the inferotemporal (IT) cortex:

  • V4 (Curved Contour Cells): These units detect local curvature and corners.
  • PIT (Convex Cells): Located in the posterior inferotemporal cortex, these cells have annulus-shaped (ring-like) receptive fields (RFs). They fire most strongly when centered on a figure, effectively "probing" for the medial axis.
  • AIT (Teardrop Cells): In the anterior inferotemporal cortex, these units integrate signals from Convex Cells across different scales and positions.
The model uses excitatory and inhibitory connections between active and inactive on-surround units to resolve figure-ground ambiguity.

The model's core innovation is the feedback loop between AIT and PIT. If AIT units detect "closure" (indicating a point is inside a boundary), they send excitatory feedback to enhance the interior signal. Conversely, if no closure is detected—such as in the concave "mouth" of a 'C'—the signal is suppressed.

Key Insights: Jitter and the Medial Axis

A central feature of the model is its treatment of biological "noise." Rather than treating the variation in neuron RF sizes as "noise," the model uses it as a multi-scale tool to find the medial axis (the "skeleton") of any shape, regardless of its thickness or size.

Jitter in receptive field sizes and locations allows the model to sample the stimulus at multiple scales, aiding in the detection of the medial axis.

By grouping these varied signals, Teardrop cells can detect closure and anchor neural activity to the shape's skeleton. This provides a compact, efficient way for the brain to represent object geometry, suggesting that "interior enhancement" is a peak of activity along the medial axis rather than a flat fill.

Key Results and Performance

The model was evaluated using several types of visual stimuli, including standard geometric shapes, texture-defined squares, and 1,500 algorithmically generated "random block" shapes. Performance was measured using three indices: the In-Out Index (IOI), Medial Axis Index (MAI), and Boundary Index (BI).

The "Feedback-Intact" model significantly outperformed "No-Feedback" versions, especially as shape complexity increased. Furthermore, the model successfully replicated the 80–100ms delay seen in biological "late component" responses, where initial activity represents edge detection and subsequent sustained activity represents feedback-driven interior enhancement. The model also maintained size invariance and correctly identified the medial axis in natural images from the Berkeley Segmentation Dataset.

Limitations and Future Work

While the teardrop model provides a robust framework, it currently simplifies early vision (Retina to V1) into a single edge-map stage. Future iterations could model the specific contributions of V1/V2 horizontal connections and address how the brain handles occlusion or amodal completion. Additionally, while "Teardrop cells" are a strong theoretical prediction, their exact physiological counterparts in the anterior inferotemporal cortex remain to be explicitly identified in vivo.

Summary updated

Abstract

Determining whether a region belongs to the interior or exterior of a shape (figure-ground segregation) is a core competency of the primate brain, yet the underlying mechanisms are not well understood. Many models assume that figure-ground segregation occurs by assembling progressively more complex representations through feedforward connections, with feedback playing only a modulatory role. We present a dynamical model of figure-ground segregation in the primate ventral stream wherein feedback plays a crucial role in disambiguating a figure's interior and exterior. We introduce a processing strategy whereby jitter in RF center locations and variation in RF sizes is exploited to enhance and suppress neural activity inside and outside of figures, respectively. Feedforward projections emanate from units that model cells in V4 known to respond to the curvature of boundary contours (curved contour cells), and feedback projections from units predicted to exist in IT that strategically group neurons with different RF sizes and RF center locations (teardrop cells). Neurons (convex cells) that preferentially respond when centered on a figure dynamically balance feedforward (bottom-up) information and feedback from higher visual areas. The activation is enhanced when an interior portion of a figure is in the RF via feedback from units that detect closure in the boundary contours of a figure. Our model produces maximal activity along the medial axis of well-known figures with and without concavities, and inside algorithmically generated shapes. Our results suggest that the dynamic balancing of feedforward signals with the specific feedback mechanisms proposed by the model is crucial for figure-ground segregation.
BibTeX
@inproceedings{layton2014neural,
  author = {Oliver W. Layton and Ennio Mingolla and Arash Yazdanbakhsh},
  title = {{Neural dynamics of feedforward and feedback processing in figure-ground segregation}},
  booktitle = {Frontiers in Psychology},
  year = {2014},
  doi = {10.3389/fpsyg.2014.00972},
}
Figure