An adaptive person recognition system

We propose a framework for integrating the processes of object recognition and knowledge adaptation. This framework acknowledges that the performance of the object recognition process depends directly on the state of the system's internal knowledge, i.e., its memory, and conversely, that the efficacy of the system's knowledge adaptation process is enhanced by its ability to recognize objects with greater accuracy. Thus, object recognition and knowledge adaptation are inseparable aspects of the same cognitive task, and must be co-ordinated if the system is to be robust in the context of objects that present variations in illumination, scale, shift, and other temporal variations. Specifically, the presented system combines a multiple-cue person recognition component and an example-based knowledge adaptation component, and is applied to the task of automatic video indexing of personal appearance events. We present the details of this integrated framework and demonstrate successful experimental results.