Adaptive semi-supervised spectral clustering based on Nyström method

In this paper, we proposed AS³C-N algorithm, a method of adaptive semi-supervised spectral clustering based on Nyström approximation, and apply it to color image classification. Firstly, Introduction and analysis of spectral grouping using the Nyström method are given. compared with NJW spectral clustering, Nyström approximation can reduce the requirements for computer time and space; Secondly, we can use the pairwise constraints information in the spectral clustering as background prior knowledge for semi-supervised learning; At last, we proposed an method that can select the scaling parameter in computing affinity matrix automatically, avoiding selecting σby running clustering algorithm repeatedly and improving the instability of classification result from the selected samples randomly. Experiment results from image classification shows that AS³C-N performs better than spectral clustering with fixed scaling parameter, and it can improve the classification accuracy, especially in the fuzzy boundary. That may be feasibility and effectiveness in dealing with the practical problems.