A Fast Large Size Image Segmentation Algorithm Based on Spectral Clustering

As spectral clustering has the advantage of recognizing non-convex distribution, it has been widely used in image segmentation and other areas. However, when the spectral clustering algorithm deals with large size images, in order to get the affinity matrix, it costs a lot of computation time, so its applications are limited. To solve this problem, this paper proposes a fast image segmentation algorithm based on spectral clustering. Firstly, we separate the large size image into several smaller images which are segmented in advance, and combine the segmentation results of each smaller image. Then a point is randomly selected in the integrated results to constitute the feature data of the large size image. The feature data is clustered to get the image segmentation results by using the spectral clustering method. Finally, we show the effectiveness of the proposed algorithm by experiments and compare with the common used image segmentation method.