A data-driven approach to stimulus selection reveals an image-based representation of objects in high-level visual...

Area of Science:

• Neuroscience
• Cognitive Science
• Computer Vision

Background:

• The ventral visual pathway is crucial for object recognition.
• Understanding whether neural object representations rely on low-level or high-level properties is an ongoing challenge.
• Experimental limitations in object sampling can bias findings on neural representations.

Purpose of the Study:

• To investigate the role of image properties in object representation within the ventral visual pathway.
• To develop a data-driven method for stimulus selection to overcome sampling biases.
• To determine if similar image properties elicit similar neural responses, irrespective of semantic category.

Main Methods:

• A data-driven approach was used to describe and select objects based on multi-dimensional image properties.
• Objects were sampled evenly from this image space in the first experiment.
• In the second experiment, object clusters were selected for their image property similarity to specific object categories (bottle, chair, face, house, shoe).

Main Results:

• Distinct patterns of neural response were observed even for object clusters lacking consistent semantic features.
• Object clusters with image properties similar to a category elicited neural response patterns comparable to that category.
• Neural response patterns were consistent across participants, suggesting common organizational principles.

Conclusions:

• Object perception in the ventral visual pathway is significantly influenced by image properties.
• A data-driven, image-property-based stimulus selection offers a more ecologically valid approach to studying neural representations.
• Common organizational principles in object representation are evident across individuals.