Human-guided fuzzy decision for image similarity analysis and classification based on information compression

Abstract

This paper introduces an original unsupervised learning algorithm for information compression that is further used in the proposed fuzzy inference procedure for discovering similarities between different images for the purpose of their classification. Two features extracted from each compressed information model are used in the paper to represent the location of the compressed model in the three-dimensional red-green-blue (RGB) space and its size (volume). A method for tuning the fuzzy inference procedure is proposed in the paper that uses a predefined human preference in the form of a given list of similar images with their approximate similarity levels. Thus the whole computation scheme is a kind of human-guided similarity analysis. The choice of the optimization algorithm and the selection of the optimization criterion are among the important problems, discussed in the paper. The final goal is to achieve a plausible “human-like'' decision for similarity, when processing large number of images and other pictorial information. The whole proposed computation scheme for similarity analysis and classification is illustrated on a test example of flower images followed by detailed discussions. © 2010 Wiley Periodicals, Inc. © 2011 Wiley Periodicals, Inc.