Cue-guided search: a computational model of selective attention

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Computer Vision

Background:

• Selective visual attention integrates external stimuli with internal task-specific knowledge.
• Understanding this interaction is key to how we process complex scenes.
• Existing models often focus on saliency, neglecting dynamic top-down influences.

Purpose of the Study:

• To propose a novel computational model for selective visual attention in natural environments.
• To model the dynamic integration of bottom-up and top-down information for visual search.
• To demonstrate the model's effectiveness in a face detection task.

Main Methods:

• Developed a computational model integrating bottom-up visual features and top-down task-relevant cues.
• Utilized a novel interactive spiking neural network (ISNN) with a specific activation rule.
• Implemented learning rules for bottom-up and top-down weight parameters.
• Applied the model to a cued face detection task.

Main Results:

• The model successfully integrates bottom-up and top-down information to guide attention.
• Attention trajectories were significantly influenced by the interaction of information and cue variations.
• The model produced appropriate, task-relevant search patterns in the face detection task.

Conclusions:

• The proposed model effectively simulates selective visual attention by integrating bottom-up and top-down processing.
• The results align with psychological evidence on attentional mechanisms.
• This approach offers a more comprehensive understanding of visual search in naturalistic settings.