Towards Fuzzy Classification in Cbir

At present a great deal of research is being done in different aspects of Content-Based Image Retrieval (CBIR). Image classification is one of the most important tasks that must be dealt with in image DB as an intermediate stage prior to further image retrieval. The issue we address is an evolution from the simplest to more complicated classifiers. Firstly, there is the most intuitive one based on a comparison of the features of a classified object with a class pattern. We propose a solution to the problem of finding the adequate weights, especially in the case of comparing complex values of some features. Secondly, the paper presents decision trees as another option in a great number of classifying methods. Thirdly, to assign the most ambiguous objects we have built fuzzy rule-based classifiers. We propose how to find the ranges of membership functions for linguistic values for fuzzy rule-based classifiers according to crisp attributes. In this paper, we present the promising results of the three above-mentioned classifications. Experiments demonstrate the precision of each classifier for the crisp image data in our CBIR. Furthermore, these results are used to construct a search engine, taking into account data mining. If the classification precision appears insufficient for the search engine requirements, in the next step fuzzy decision trees will be introduced.