To What Extent Does Global Shape Influence Category Representation in the Brain?

N. Apurva Ratan Murty, R. T. PramodJournal of Neuroscience 2016 · DOI · OpenAlex

Abstract

Humans are remarkably good at recognizing and rapidly categorizing objects into semantic categories (such as faces, cars, animals, etc.), and this ability is thought to arise out of a series of increasingly complex hierarchical computations in the visual cortex. Visual information from the retina is processed through a hierarchy of cortical areas, yet a central debate remains: does the brain group objects based on what they are (semantics) or what they look like (global shape)? This paper reviews evidence from Bracci and Op de Beeck (2016), who decoupled shape from category using fMRI and Representational Similarity Analysis (RSA). The findings suggest that high-level brain regions represent semantic categories independently of global shape, with the ventral pathway emphasizing animacy and the dorsal pathway focusing on action-relatedness.

Introduction

In visual neuroscience, a long-standing "chicken and egg" problem persists regarding object categorization. While specific regions in the ventral visual stream—the "what" pathway—respond selectively to categories like faces or tools, objects within these categories often share similar physical geometries. For instance, most animals possess heads and limbs, while many tools feature elongated handles. This correlation makes it difficult to determine if neural responses are driven by semantic meaning (e.g., "this is a living thing") or merely by shared visual geometry (global shape).

To address this, Bracci and Op de Beeck (2016) designed a study to break the correlation between shape and category, investigating whether category representation exists as a distinct organizational principle in the human brain.

Methodology

The researchers employed a stimulus design specifically engineered for Representational Similarity Analysis (RSA) to isolate the effects of shape and category.

Stimulus Uncoupling

The study utilized a custom-curated dataset of 54 naturalistic images spanning six categories: minerals, animals, fruits/vegetables, musical instruments, sporting equipment, and tools. Crucially, the researchers selected nine distinct global shapes and ensured each category contained one exemplar for every shape. For example, a "round" animal (a coiled snake) and a "round" tool (a circular saw) shared a global silhouette but belonged to different semantic categories. This design ensured that silhouette shape and behavioral category were not correlated.

Behavioral and Neural Mapping

Subjects participated in behavioral tasks to arrange these stimuli based on either shape similarity or category similarity, creating behavioral dissimilarity matrices. During fMRI scanning, neural activity was measured across 15 regions of interest (ROIs) in both the ventral (object recognition) and dorsal (action/spatial) pathways. The researchers then correlated these neural patterns with the behavioral models to determine which factor—shape or category—best explained the activity in each region.

Key Insights

The Posterior-to-Anterior Gradient

The study revealed a clear hierarchy of abstraction within the ventral stream. Posterior regions (located toward the back of the brain) appeared to be "slaves to the pixels," primarily representing low-level silhouette shapes. However, as processing moved anteriorly (forward) toward the temporal lobe, the neural representation of raw shape diminished in favor of semantic category. This suggests that the brain progressively discards geometric detail to extract categorical meaning.

Functional Divergence of Pathways

A striking finding was the divergence in how categories are organized across the two primary visual streams:

  • Ventral Pathway: This stream is primarily organized by Animacy (distinguishing animate from inanimate objects).
  • Dorsal Pathway: This stream is organized by Action (distinguishing tools and action-related objects from non-action objects).

Shape-Category Integration

In certain high-level regions, both "perceptual shape" (how humans perceive similarity) and "category" explained neural patterns. This suggests that at advanced stages of processing, the brain moves beyond simple geometry toward a more complex "perceptual shape" that begins to merge with semantic identity.

Results and Discussion

The results confirmed that in anterior ventral regions, category information remains significant even after statistically controlling for global shape. Hierarchical clustering of neural data showed that the anterior ventral stream clearly separates animate from inanimate objects, whereas dorsal regions (parietal and frontal) separate objects based on whether they are manipulated by hand.

While this study provides strong evidence for independent category representation, some limitations remain. The control for global shape does not necessarily account for local features or textures (e.g., the texture of fur vs. metal). It is possible that what appears to be a "category" response is a response to these fine-grained visual textures. Furthermore, while the study used pre-defined ROIs, subsequent searchlight analyses have confirmed these patterns hold across the wider cortex. Future research using high-temporal resolution techniques like EEG or MEG will be necessary to determine the timing of these signals and whether visual features precede or follow categorical assignment.

Conclusion

By decoupling shape from category, this research demonstrates that the human brain possesses specialized modules for semantic categorization that function independently of global geometry. The distinct organization of the ventral and dorsal pathways—focusing on animacy and action, respectively—highlights how the brain's representational structure is optimized for different functional requirements of human cognition.

Summary updated

Abstract

Humans are remarkably good at recognizing and rapidly categorizing objects into semantic categories (such as faces, cars, animals, etc), and this ability is thought to arise out of a series of increasingly complex hierarchical computations in the visual cortex. Visual information from the retina is
BibTeX
@article{murty2016what,
  author = {N. Apurva Ratan Murty and R. T. Pramod},
  title = {{To What Extent Does Global Shape Influence Category Representation in the Brain?}},
  journal = {Journal of Neuroscience},
  year = {2016},
  doi = {10.1523/jneurosci.0387-16.2016},
}
Figure