Combining the self-adaptive neural network and support vector machine for online clustering and image segmentation

The difficulties of online clustering are how to handle variation of cluster number in the same framework, how to be computationally efficient for real time applications, and how to ensure the error convergence of the algorithm. This paper presents a new online clustering algorithm that combines the self-adaptive neural network and support vector machine, which is used to learn continuously evolving clusters from non-stationary data. The online clustering algorithm uses a fast adaptive learning procedure to take into account variations over time. In non-stationary and multi-class environment, the algorithm learning procedure consists of five main stages: creation, adaptation, mergence, split and elimination. One of limitations of existing segmentation algorithms is that these algorithms cannot adapt to real-world changes. The proposed algorithm may solve the problem, and uses the proposed algorithm to do image segmentation. Experiments are carried out to illustrate the performance of the proposed algorithm. Compared with SAKM algorithm, our algorithm shows better performance in accuracy of clustering. On Berkeley image data set, we do image segmentation to compare our algorithm with Nyström method, and results show our algorithm had better performance. On the timevarying meteorological satellite FY-2 water vapor images, we further test our algorithm for image segmentation.