A Fuzzy Approach to Image Texture Representation Applied to Visual Coarseness Description

Although "texture" is one of the most used features in image analysis, it is an ambiguous concept which, in many cases, is not easy to characterize. In this paper we face the problem of imprecision in texture description by proposing a methodology to represent texture concepts by means of fuzzy sets. Specifically, we model the concept of "coarseness", the most extended in texture analysis, relating representative measures of this kind of texture (usually some statistic) with its presence degree. To obtain these "presence degrees" related to human perception, we propose a methodology to collect assessments from polls filled by human subjects, performing an aggregation of these assessments by means of OWA operators. Using as reference set a combination of some statistics, the membership function corresponding to the fuzzy set "coarseness" will be modelled as the function which provides the best fit of the collected data. The proposed methodology could be extended to other types of texture concepts like orientation, roughness or regularity. The main novelty of this approach is the introduction of semantics in the texture analysis problem, by using linguistic labels represented by fuzzy sets in order to describe texture features.