Recognition of indoor images employing supporting relation between objects

In this paper, we describe a new design of a recognition system for a single image of an indoor scene including complex occlusions. In conventional works, the systems could not recognize images of an indoor scene including complex occlusions. Our system can treat them by employing supporting relation between objects. In our system, first, the system estimates the 3D structure of an object by fitting a 3D structure model to the image qualitatively. Next, by checking the supporting relation between objects, it eliminates object candidates that cannot exist and estimates real objects from their parts in the image. Finally, the system recognizes objects that are compatible with each other. We implemented the system as a multi-agent-based image understanding system. In this paper, we describe the design of the system and results of experiments.