Spectral Clustering on Multispectral Images Using Mean Shift Based Similarity Learning
5 Pages Posted: 26 Apr 2019
Date Written: April 26, 2019
Abstract
Spectral clustering is a popular graph-based clustering technique. There is difficulty in similarity graph construction and curse of dimensionality is complex to deal with general spectral clustering algorithm consists of three steps, i) similarity graph construction, ii) Eigenvectors and Eigenvalues evaluation and iii) clustering with k means. This paper proposes a modified spectral algorithm that uses a similarity graph using Mean shift based Anchor graph, Eigenvectors, and Eigenvalues evaluation based on singular value decomposition, instead of K-Means algorithm with MeanShift clustering. Experimentation on multispectral images using modified spectral clustering shows improvement in the clustering accuracy and can resolve the dimensionality problem when compared with traditional techniques.
Suggested Citation: Suggested Citation