Texture-like representation of objects in human visual cortex.

Area of Science:

• Neuroscience
• Computer Vision
• Visual Perception

Background:

• Object recognition is crucial for human vision.
• Category-selective regions in the visual cortex are implicated in object recognition.
• Current models debate whether these regions represent objects directly or use general visual features.

Purpose of the Study:

• To investigate whether category-selective visual cortex represents objects or visual textures.
• To differentiate between object-specific and texture-based representations in visual processing.

Main Methods:

• Utilized a novel image synthesis technique for independent control of feature complexity and spatial arrangement.
• Compared human observer performance with computational models (BOLD responses, macaque IT cortex models, deep convolutional neural networks) on object detection tasks.

Main Results:

• Humans could easily distinguish natural objects from spatially scrambled synthetic images with similar features.
• Computational models, including those based on category-selective regions, failed to identify the natural object.
• Observer models accurately predicted human performance in categorization, indicating sufficient signal was present.

Conclusions:

• Category-selective visual cortex likely provides a basis set of texture-like features, not explicit object representations.
• These texture features are reconfigurable, enabling flexible learning and identification of novel object categories.
• The findings challenge traditional views of object-specific coding in visual cortex.