An image understanding system for various images based on multi-agent architecture

An image understanding system for real world images which has an ability to recognize various kinds of images is proposed. We propose a multi-agent architecture to integrate and cooperate object recognition modules for individual target objects. In our system, object candidates generated by different agents are integrated not only on the evaluations by each modules themselves but also on spatial relations among objects. By checking spatial relation, the agents also estimate actual objects from parts seen in the image. Such mechanisms are realized by autonomous cooperation among the agents, and the most reliable result is selected after the arbitration between them. We implemented an experimental system on a PC cluster system, and achieved recognition for both indoor and outdoor images.