Granular Computing in the Information Transformation of Pattern Recognition

In the past decade, many papers about granular computing(GrC) have been published, but the key points about granular computing(GrC) are still unclear. In this paper, we try to find the key points of GrC in the information transformation of the pattern recognition. The information similarity is the main point in the original insight of granular computing (GrC) proposed by Zadeh(1997[1]). Many GrC researches are based on equivalence relation or more generally tolerance relation, equivalence relation or tolerance relation can be described by some distance functions and GrC can be geometrically defined in a framework of multiscale covering, at other hand, the information transformation in the pattern recognition can be abstracted as a topological transformation in a feature information space, so topological theory can be used to study GrC. The key points of GrC are (1) there are two granular computing approaches to change a high dimensional complex distribution domain to a low dimensional and simple domain, (2) these two kind approaches can be used in turn if feature vector itself can be arranged in a granular way.