Model-driven interpretation in intelligent vision systems

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Computer Vision

Background:

• Current psychological theories of perception lack comprehensive accounts.
• Artificial intelligence (AI) programs for scene analysis offer potential insights.
• Existing AI models for object recognition have limitations.

Purpose of the Study:

• To review and critique AI scene analysis programs.
• To explore how AI can inform psychological theories of perception.
• To propose new methods for incorporating object knowledge in AI.

Main Methods:

• Critique of existing AI scene analysis programs.
• Analysis of methods for incorporating a priori knowledge using models.
• Development of criteria for evaluating model information.
• Contrast of various AI proposals.

Main Results:

• AI scene analysis programs demonstrate potential for enhancing psychological perception theories.
• Exploration of diverse methods for integrating object models in AI.
• Identification of limitations in current AI approaches to scene analysis.

Conclusions:

• AI scene analysis, particularly through model integration, offers a promising avenue for advancing perception research.
• Two novel proposals for exploiting model information are presented, building upon the POLY program.
• Further development of AI models can lead to more adequate accounts of human perception.