An Image-Segmentation Method Based on Improved Spectral Clustering Algorithm

An image-segmentation method based on the improved spectral clustering algorithm is proposed. Firstly, the improved hill-climbing is applied to searching for the optimal clustering center in order to avoid classical spectral clustering algorithm’s heavy dependency on initial clustering center. The search directions are added to improve global search capability of hill-climbing method to avoid the local optimum. Secondly, to reduce cost of computation, the pixels that have the same gray scale values are merged in the image. Finally, the improved spectral clustering algorithm is applied to image-segmentation. The experiment results prove that the method proposed in the paper is more stable, fast and effective, and the effect of the segmentation is better and obvious.