From Rough through Fuzzy to Crisp Concepts: Case Study on Image Color Temperature Description

Summary

This chapter proposes a framework for designing interval-based classifiers for fuzzy categories based on rough information systems. The information system is given by joining objective measurements of a quantized scalar featuretfor a class of objects with subjective decisions (votes) regarding the categoryc,to which objects belong. Using the roughness degree, we estimate at sparse points unknown fuzzy membership functions forncategories. Having such sparse membership functions and the original information system, we find a family of optimal partitions of a feature’s range for two important cost functions. In practice, only interval partitions are useful for fast decisions. The algorithm generating optimal intervals is given and applied to extracting of image color temperature descriptions.