抄録
In this paper, we introduce a strategy which chooses significant features based on information theory in 3-D object recognition. Our optimality criterion is a reduction of uncertainness in a recognition process. If uncertainty and ambiguity of the recognition process can be reduced, object recognition becomes more reliable. A technique to choose the optimal feature based on information theory is already studied for active object recognition. This paper proposes a feature selection strategy for recognizing 3-D objects by extending such a framework. The strategy is constructed by an entropy-based approach using an iterative algorithm. Significant features is chosen based on a set of geometric features consisting of three features in the feature selection strategy. We present 3-D object recognition process, and discuss the validity of the proposed feature selection strategy via some experiments.
