Extension of Mean Shift vector with theoretical analysis and experiment

Abstract

Mean shift algorithm is a statistics iterative algorithm which is widely used, its increment (namely mean shift vector) of iterative point in each iteration step changes adaptively. This paper presents an extensional mean shift vector, and proves convergence of mean shift algorithm which using the extensional mean shift vector. In addition, we did an experiment - using mean shift algorithm to solve the local Maximum of kernel-based density estimation, in our experiment, the convergence rate of mean shift algorithm which using extensional mean shift vector reach twice the convergence rate of mean shift algorithm which using traditional mean shift vector.