Realtime parallel and cooperative recognition of human face for a naturalistic visual human interface

The concept of a Visual Software Agent (VSA), i.e., an anthropomorphic agent possessing a realistic human form and communication functions that create an interface with the user, is proposed. Research is proceeding on its implementation on a parallel computer. In this paper, we describe a method that is effective as the visual input function of the anthropomorphic agent. This method recognizes and extracts only one person (user) directly facing the camera in a moving image in which several people occasionally enter and leave. The recognition method can be executed in real time in a typical indoor environment. The background and the lighting do not have to be restricted and recognition can be achieved in real time by a parallel computer. In addition, this technique simulates the initial stage of human visual functions and is based on the parallel extraction and integration of multiple image features. The movements of visual attention of cognitive psychology can also be simulated. An anthropomorphic agent derived from this approach successively looks at multiple human faces, fixes its gaze only on a specific person, and synthesizes agent behavior that approximate that of humans. These moving image recognition and synthesis processes were implemented on a newly designed parallel computer.