Visual crowding and category specific deficits for pictorial stimuli: A neural network model

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Previous models often lack unsupervised components for visual processing.
• Real pictorial stimuli and diverse object categories are crucial for realistic simulations.
• Understanding category-specific recognition failures is key to cognitive models.

Purpose of the Study:

• To simulate visual object processing using a modular neural network.
• To investigate the role of supervised and unsupervised modules in object recognition.
• To explore the causes of category-specific recognition impairments.

Main Methods:

• Developed a modular neural network with unsupervised and supervised components.
• Processed real pictorial representations of various object categories.
• Assessed model performance via learning, generalization, and simulated lesion damage.

Main Results:

• Observed significant category effects, with living things and musical instruments showing higher recognition failure rates.
• Identified within-category similarity and visual crowding as factors contributing to impairments.
• Demonstrated that increased competition between representations leads to category-specific deficits.

Conclusions:

• The model supports the cascade model of object recognition.
• Within-category similarity and visual crowding are critical for understanding category-specific impairments.
• This modular approach provides insights into visual object processing and recognition failures.