On some convergence properties of the subspace constrained mean shift

作者：

Highlights：

• We investigate convergence properties of the MS and SCMS algorithms.

• The SCMS is shown to inherit important convergence properties of the MS algorithm.

• Theoretical guarantees for stopping criteria are provided.

• Three variations of the SCMS are proposed and tested.

摘要

Highlights•We investigate convergence properties of the MS and SCMS algorithms.•The SCMS is shown to inherit important convergence properties of the MS algorithm.•Theoretical guarantees for stopping criteria are provided.•Three variations of the SCMS are proposed and tested.

论文关键词：Unsupervised learning,Subspace constrained mean shift,Dimensionality reduction,Principal curves,Principal surfaces,Convergence

论文评审过程：Received 8 January 2013, Revised 8 April 2013, Accepted 23 April 2013, Available online 30 April 2013.

论文官网地址：https://doi.org/10.1016/j.patcog.2013.04.014