On fuzzy partitions for visual texture modelling

Texture is one of the most used low-level feature for image analysis and, in addition, one of the most difficult to characterize due to its imprecision. It is usual for humans to describe visual textures according to some perceptual properties like coarseness-fineness, orientation or regularity. In this paper, we propose to model the fineness property, that is the most popular one, by means of a fuzzy partition on the domain of representative fineness measures. In our study, a wide variety of measures is studied, and the partitions are obtained by relating each measure (our reference set) with the human perception of fineness. Assessments about the perception of this property are collected from pools. This information is used to analyze the capability of each measure to discriminate different fineness categories, which imposes the number of fuzzy sets of the partition. Moreover, it is used to calculate the parameters of the membership function associated to each fuzzy set.