A similarity measure of complex-valued evidence theory for multi-source information fusion

Abstract

The study of complex-valued evidence theory has become an interesting topic, particularly in the context of information fusion techniques. These studies mainly focus on its geometric interpretation and application. Moreover, little work has been conducted with regard to conflict measure. Therefore, this paper defines a complex similarity measure which quantifies the degree of conflict among complex-valued mass functions by measuring the degree of consistency among complex-valued mass functions. It also satisfies some basic properties, such as non-negativity, symmetry and etc. Complex similarity measure is an extension of traditional similarity measure in complex space, which will collapse to a similarity measure of real space when complex-valued mass functions collapse to mass functions of real space. Furthermore, some numerical examples illustrate the rationality and effectiveness of complex similarity measure. By employing complex similarity measure, this paper proposes a method to combine information and reduce the time cost associated with complex-valued Dempster's rule of combination. Experimental results on datasets indicate that for some data sets (Take iris for example), the proposed method can not only reduce the time cost (75%), but also improve the accuracy of fusion results (96%).