Improved Mean Shift Spectral Clustering Based on Reduced Set Density Estimator

Author

Qian, Pengjiang ; Wang, Shitong ; Wu, Xiaojun ; Deng, Zhaohong ; Sang, Qingbing

Author_Institution

Sch. of Inf. Technol., Jiangnan Univ., Wuxi, China

fYear

2009

fDate

4-6 Nov. 2009

Firstpage

1

Lastpage

5

Abstract

Mean shift spectral clustering (MSSC) brings us an alternative for image segmentation. However, owing to being based on the classical Parzen window estimator (PW) and employing the full data sample for density estimation, the usefulness of MSSC is weakened. In this paper, the improved mean shift spectral clustering (IMSSC) algorithm is proposed by replacing PW with the reduced set density estimator (RSDE). Due to just a few sample points in the reduced set being referred to, the time complexity of mean shift embedded in IMSSC decreases to O(mN) and the total computational costs of IMSSC are sharply reduced.

Keywords

computational complexity; estimation theory; image segmentation; pattern clustering; image segmentation; improved mean shift spectral clustering algorithm; reduced set density estimator; time complexity; Clustering algorithms; Computational efficiency; Constraint optimization; Convergence; Image segmentation; Information technology; Kernel; Minimization methods; Probability density function; Testing;

fLanguage

English

Publisher

ieee

Conference_Titel

Pattern Recognition, 2009. CCPR 2009. Chinese Conference on

Conference_Location

Nanjing

Print_ISBN

978-1-4244-4199-0

Type

conf

DOI

10.1109/CCPR.2009.5343985

Filename

5343985