Top-down attention based on object representation and incremental memory for knowledge building and inference.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Cognitive Science

Background:

• Human visual perception relies on incremental learning and selective attention for object recognition.
• Existing models often lack efficient mechanisms for real-time object localization and recognition of novel objects.

Purpose of the Study:

• To propose a task-specific top-down attention model for object localization using form and color.
• To integrate bottom-up saliency with top-down attentional mechanisms for enhanced visual perception.
• To develop an incremental learning system capable of memorizing and inferring object features.

Main Methods:

• A novel selective attention model incorporating top-down bias signals for target form and color features.
• Integration of bottom-up saliency based on primitive visual features and memory modules.
• Utilizing a Growing Fuzzy Topology Adaptive Resonance Theory (GFTART) network for incremental learning, feature memorization, and top-down bias generation.

Main Results:

• The proposed model successfully focuses on specified target objects in natural scenes.
• Demonstrated incremental representation and memorization of diverse object features.
• The model effectively infers new, unknown objects based on learned form and color features.

Conclusions:

• The developed attention model enhances object recognition through efficient selective attention and incremental learning.
• The GFTART network facilitates robust object feature learning, memorization, and inference capabilities.
• This approach offers a promising pathway for developing more sophisticated artificial visual perception systems.