Title: Median relative intersection of confidence intervals for bandwidth estimation in mean shift clustering technique

Authors: Prasad Kaviti; Valli Kumari Vatsavayi

Addresses: Department of Computer Science and Systems Engineering, AU College of Engineering, Andhra University, AP, India ' Department of Computer Science and Systems Engineering, AU College of Engineering, Andhra University, AP, India

Abstract: Mean shift algorithm is a non-parametric iterative algorithm widely used in segmentation, clustering, object tracking, etc. However, tuning the bandwidth parameter and selection of kernel with its convergence is required. This paper proposes a modified mean shift in terms of bandwidth selection and its adequate kernel selection. Mean shift equipped with median relative intersection of confidence intervals (MRICI) for multispectral image clustering is proposed. Initially different kinds of bandwidth estimators like static, Silverman, Scott, ICI and MRICI are evaluated and are considered four classes of kernels Gaussian, Epaenchnikov, flat, biweight with general convergence. Later different combinations of the four classes of kernels and different bandwidth estimators of mean shift are evaluated. Results show an improvement in intracluster similarity based on silhouette measure for MRICI bandwidth estimation using the Gaussian kernel of mean shift when compared to other combinations of mean shifts.

Keywords: mean shift clustering; kernels; bandwidth; confidence intervals; multispectral images.

DOI: 10.1504/IJAIP.2025.149742

International Journal of Advanced Intelligence Paradigms, 2025 Vol.30 No.5, pp.351 - 370

Received: 30 Nov 2018
Accepted: 05 Feb 2019
Published online: 12 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article