In pattern recognition, understanding the structure of pattern data is important for measuring sufficiency of features at hand and for designing classifiers. We have proposed a visualization method with the help of graph representation in which discriminant information in the original space is correctly reflected. This method is different from conventional methods in the following point: Instead of individual data points, some subsets of data points are used in the proposed method in such a way that we construct a graph in which subsets are shown as nodes and the intersections among these subsets are shown as edges. In this paper, the proposed method is compared with some conventional visualization methods in order to make the characteristics of these methods clear and to confirm the effectiveness of the proposed method.
展开▼
