Customizing fuzzy partitions for visual texture representation

Abstract

Visual textures in images are usually described by humans using linguistic terms related to their perceptual properties, like “very coarse”, “low directional”, or “high contrasted”. Thus, computational models with the ability of providing a perceptual texture characterization on the basis of these terms play a fundamental role in tasks where some interaction with subjects is needed. In this sense, fuzzy partitions defined on the domain of computational measures of the corresponding property have been proposed in the literature. However, the main drawback of these proposals is that they do not take into account the subjectivity associated to the human perception. For example, the perception of a texture property may change depending on the user, and in addition, the image context may influence the global perception of the properties. In this paper, we propose to solve these problems by means of a methodology that automatically adapts any generic fuzzy partition modeling a texture property to the particular perception of a user or to the image context. In this method, the membership functions associated to the fuzzy sets are automatically adapted by means of a functional transformation on the basis of the new perception. For this purpose, the information given by the user or extracted from the textures present in the image are employed.