Spectral clustering algorithm combining local covariance matrix with normalization

• Published in: Neural computing & applications Vol. 32; no. 11; pp. 6611 - 6618
• Main Authors: Du, Tingting, Wen, Guoqiu, Cai, Zhiguo, Zheng, Wei, Tan, Malong, Li, Yangding
• Format: Journal Article
• Language: English
• Published: London Springer London 01.06.2020; Springer Nature B.V
• Subjects: Affinity; Algorithms; Artificial Intelligence; Clustering; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Covariance matrix; Data Mining and Knowledge Discovery; Image Processing and Computer Vision; Multi-Source Data Understanding (MSDU); Probability and Statistics in Computer Science; Spectra; Clusters intersection; Local covariance; Spectral clustering; Affinity matrix learning
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-018-3852-z

Affinity matrix construction is a key step in the spectral clustering. However, traditional spectral clustering methods usually ignore the intersection problem that may exist between the different clusters of data, so the resulting matrix could be unreliable. This paper proposes a new local covariance-based method to solve the above problem. Specifically, we first learn an initial affinity matrix by adding the local covariance into traditional matrix construction step, which could guarantee the obtained matrix avoids the impact of the intersection point while preserving the neighborhood relationship of data. We then employ the normalized Laplacian on the obtained matrix to further improve the clustering performance. The ACC and NMI of the proposed method increased by 6.40% and 5.33% on average compared with six classical spectral clustering methods. Experimental evaluation on eight benchmark data sets shows that the proposed method has better clustering performance.