Adaptive Selection of Image Classifiers

Recently, the concept of \Multiple Classi er Systems" was proposed as a new approach to the development of high performance image classi cation systems. Multiple Classi er Systems can be used to improve classi cation accuracy by combining the outputs of classi ers making \uncorrelated" errors. Unfortunately, in real image recognition problems, it may be very di cult to design an ensemble of classi ers that satis es this assumption. In this paper, we propose a di erent approach based on the concept of \adaptive selection" of multiple classi ers in order to select the most appropriate classi er for each input pattern. We point out that adaptive selection does not require the assumption of uncorrelated errors, thus simplifying the choice of classi ers forming a Multiple Classi er System. Reported results on the classi cation of remote-sensing images show that adaptive selection can be used to obtain substantial improvements in classi cation accuracy.