Fuzzy Sets for Modelling Fineness Perception in Texture Images

Fineness is a primary texture feature frequently used for image content description. However, it is an ambiguous concept difficult to be characterized. To face this ambiguity, we propose to model the fineness by means of fuzzy sets, relating a representative fineness measure (our reference set) with the human perception of fineness. In our study, a wide variety of measures have been analyzed, defining a fuzzy set for each measure. The fineness perception has been collected from polls filled by human subjects, performing an aggregation of their assessments by means of OWA operators. For a given measure, the corresponding membership function is obtained by fitting the collected data. The performance of each fuzzy set is analyzed and checked with the human assessments, proposing a subgroup of them as the most adequate for modelling fineness perception in texture images.