Design of a Hybrid Classifier for Natural Textures in Images from the Bayesian and Fuzzy Paradigms

One objective for classifying textures in natural images is to achieve the best performance possible. The combination of multiple classifiers has been tested as a suitable technique. This is because the individual behaviors of the different classifiers are exploited by joining performances. Two main problems are to be addressed for combining different classifiers. First, which set of classifiers are to be selected? Second, which is the best combination method? Different strategies have been proposed focusing on both problems. We propose a new automatic approach based on two well tested classifiers, the non-parametric Parzen's windows and the Fuzzy clustering. A probability density function (PDF) is estimated through the Parzen's windows estimator to be embedded in the Bayesian framework. The decision is made by combining a likelihood value provided by the PDF and the a priori probability supplied by the Fuzzy clustering strategy through the membership degree. Experimental results for aerial images demonstrate the performance of the proposed design.