Mean shift: a robust approach toward feature space analysis

TLDR

The paper proposes a general non‑parametric technique to analyze complex multimodal feature spaces and delineate arbitrarily shaped clusters. The method uses the mean‑shift procedure as its core computational module and is applied in algorithms for discontinuity‑preserving smoothing and image segmentation that require only a resolution parameter and accept gray‑level or color images. The authors prove that the recursive mean‑shift converges to the nearest stationary point of the density, relate it to the Nadaraya‑Watson estimator and robust M‑estimators, and demonstrate its excellent performance in experiments.

Abstract

A general non-parametric technique is proposed for the analysis of a complex multimodal feature space and to delineate arbitrarily shaped clusters in it. The basic computational module of the technique is an old pattern recognition procedure: the mean shift. For discrete data, we prove the convergence of a recursive mean shift procedure to the nearest stationary point of the underlying density function and, thus, its utility in detecting the modes of the density. The relation of the mean shift procedure to the Nadaraya-Watson estimator from kernel regression and the robust M-estimators; of location is also established. Algorithms for two low-level vision tasks discontinuity-preserving smoothing and image segmentation - are described as applications. In these algorithms, the only user-set parameter is the resolution of the analysis, and either gray-level or color images are accepted as input. Extensive experimental results illustrate their excellent performance.