Multiclass mv-granular soft support vector machine: A case study in dynamic classifier selection for multispectral face recognition

This paper presents a novel formulation of multiclass support vector machine by integrating the concepts of soft labels and granular computing. The proposed multiclass mv-granular soft support vector machine uses soft labels to address the issues due to noisy and incorrectly labeled data, and granular computing to make it adaptable to data distributions both globally and locally. The proposed multiclass classifier is used for dynamic selection in a multispectral face recognition application. Specifically, for the given probe face images, mv-GSSVM is used to optimally choose one of the four options: visible spectrum face recognition, short-wave infrared face recognition, multispectral face image fusion, and multispectral match score fusion. Experimental results on a multispectral face database show that the proposed algorithm improves the verification accuracy and also decreases the computational time.