Multiscale segmentation algorithm based on subdivision and mean shift

Abstract

A fast multiscale algorithm for image segmentation is presented in this paper, which is based on mean shift skill and subdivision method. Mean shift is a nonparametric kernel density estimator, which has been applied image segmentation widely, but it can't meet the need of real-time processing. To increase the speed of segmentation of images, subdivision and its reverse skill are employed, which have applied extensively in computer aided geometric design, to convert image to different scales. The fine properties of subdivision about extraction low pass information from image lead to a very efficient and real-time nonparametric segmentation algorithm. Experiment results and further experiment data analysis show that the segmentation algorithm is faster and more practical than the mean shift algorithm and the results are satisfactory.