An Information Theoretic Approach for Active and Effective Object Recognitions

This paper describes the strategy of optimal action selection for recognition of the observed object among those registered in the image data base from an information theoretic point of view. The goal is to reduce the number and effort of the recognition steps to reach the final decision of the recognition. We have chosen an information theoretic framework, motivated by the fact that sensored image data is not noiseless or ideal, nor can the effect of a certain action be completely determined in advance. We present the strategy of the recognition process, and then, the results of the feasibility study by experiments.