Relational graph clustering based on spectral coefficient angle
Min Kong, Jin Tang, Bin Luo · 2008
This paper introduces a relational graph representation method using the angle between spectral coefficient vectors. A relational graph clustering system builds on this presentation method. The system adopts fuzzy C-mean (FCM) as clustering algorithm. FCM exerts on the pattern space which embedded by locality preserving projections (LPP). The pattern space obtains from Laplacian matrix constructed by the corner points oriented graph from image sequence. After matrix decomposition at hand, the angle between spectral coefficients vectors as spectral features are computed through eigen value and eigen vectors of it. These features can describe the distribution and relationship of all graph nodes. Experiment shows that the spectral features of the angle between spectral coefficient vectors of Laplacian graph represent image sequence correctly and graph clustering in the feature pattern space is valid.