The Neural Representation of Objects in Visual Cortex

Neuroimaging studies have shown that different categories of object evoke different neural responses in the ventral visual pathway. This has been interpreted to suggest that these regions represent high-level conceptual or semantic properties of the stimulus, such as its category. However, images from different categories differ in low-level visual properties. Therefore, the extent to which category-specific neural responses indicate high-level or low-level representations is unclear. This thesis investigates the extent to which low-level properties of objects are important in the neural response of ventral visual pathway. The first study uses a data-driven approach to select clusters of objects based on the similarity of their low-level visual properties. These visually defined clusters did not correspond to typical object categories, but still evoked distinct patterns of response in the ventral stream. The second and third studies show category-specific patterns of response in the ventral stream to scrambled objects that are not recognizable, but nevertheless retain many of their low-level visual properties. The fourth study reveals that the bias toward natural object images found in the ventral stream begins to emerge in early visual areas. The final chapter shows that category-specific patterns of EEG response can be also explained by low-level image properties. Taken together, these results demonstrate the importance of low-level visual properties in the neural representation of objects. These findings suggest that the category-selectivity observed in high-level visual regions can be explained by a distributed organization based around more basic properties of the stimulus.