A visual system theoretic cost criterion and its application to clustering and fuzzy modeling

We all know that our eyes can inherently and effectively recognize/classify objects under complex conditions. Hence, we believe that an efficient clustering approach not only depends on the principles of physical systems by which the data are generated but also on the manner that human eyes sense the structure of the data. In this study a visual system theoretic cost criterion function is proposed and based upon which a new clustering algorithm is derived. The new cost criterion is visual sampling and Weber's law is applied. The new criterion function can be made "kemelized" so that developed based on a visual system modeling of the multi-dimensional data where the visual system theories like different kernel functions can be used under different practical requirements. Furthermore, it evaluates the tightness of intra-group's data distribution and the separable degree among groups simultaneously. The experimental results demonstrate that the new clustering algorithm is especially suitable for nonlinearly separable datasets.