Abstract
The studies described here use functional magnetic resonance imaging to test whether common or distinct cognitive and/or neural mechanisms are involved in extracting object structure from the different image cues defining an object's shape, such as contours, shading, and monocular depth cues. We found overlapping activations in the lateral and ventral occipital cortex [known as the lateral occipital complex (LOC)] for objects defined by different visual cues (e.g., grayscale photographs and line drawings) when each was compared with its own scrambled-object control. In a second experiment we found a reduced response when objects were repeated, independent of whether they appeared in the same or a different format (i.e., grayscale images vs line drawings). A third experiment showed that activation in the LOC was no stronger for three-dimensional shapes defined by contours or monocular depth cues, such as occlusion, than for two-dimensional shapes, suggesting that these regions are not selectively involved in processing three-dimensional shape information. These results suggest that common regions in the LOC are involved in extracting and/or representing information about object structure from different image cues.
Introduction
How does the human brain recognize an object whether it is a detailed photograph or a simple sketch? A central question in cognitive neuroscience is whether the visual system employs specialized modules for different visual cues—such as shading or contours—or if a unified mechanism represents shape independently of its rendering. This study investigates the Lateral Occipital Complex (LOC) to determine if it acts as a "cue-invariant" hub for object recognition. By using fMRI and adaptation techniques, the researchers tested whether the LOC extracts high-level structural descriptions rather than merely processing low-level image features.
Experimental Design and Stimuli
The researchers conducted three fMRI experiments using custom-generated stimuli to isolate shape processing from semantic knowledge.
- Standard Mapping (Experiment 1): To localize the LOC, subjects viewed grayscale photographs and line drawings of novel clay objects. These were compared against "scrambled" versions that preserved low-level features (like luminance and local edges) but destroyed global shape.
- fMRI Adaptation (Experiment 2): This experiment utilized "repetition suppression." If a neural population represents an object's identity, its response should decrease (adapt) when that identity is repeated. The authors tested if adapting to a grayscale photo would reduce the response to a subsequent line drawing of the same object.
- 3D vs. 2D Comparison (Experiment 3): To see if the LOC is specifically tuned for 3D structural complexity, subjects viewed images with monocular depth cues (e.g., occlusion) versus simple 2D outlines.
Results: Evidence for Cue Invariance
The results across all three experiments point toward the LOC as a generalized shape processor.
Overlapping Activations and Adaptation
Experiment 1 revealed massive spatial overlap in the LOC for both grayscale photos and line drawings. Both formats produced significantly higher activation than their respective scrambled controls ($p < 0.001$), as demonstrated by the robust signal increases during intact object blocks (Figure 4). However, overlapping voxels do not strictly prove shared neural populations. Experiment 2 addressed this via adaptation: a significant reduction in signal was observed when object identity was repeated, even when the format changed (e.g., photo followed by drawing). This "cross-format" adaptation implies that the same neurons represent the object's geometry regardless of the visual medium.
Shape vs. Depth
Surprisingly, Experiment 3 found no significant difference in LOC activation between 3D-cued objects and 2D outlines. This suggests that while the LOC is critical for extracting global structure, it may not be the primary site for processing complex 3D depth or monocular depth cues like occlusion. Interestingly, while the LOC showed cue invariance, regions in the parietal cortex showed a preference for grayscale images, suggesting the parietal lobe might be more specialized for processing shape from specific cues like shading.
Discussion and Limitations
These findings suggest that the LOC converges on a unified representation of an object's structure from diverse visual inputs. However, limitations remain. The spatial resolution of fMRI (voxels containing thousands of neurons) means that finer-scale specialization might still exist. Additionally, while the study focused on monocular cues, previous research suggests the LOC might treat binocular depth (stereopsis) differently. Future work is needed to define the specific geometric primitives—such as parts or skeletons—that the LOC uses to code these invariant shape descriptions.
Summary updated