Dynamic image segmentation using incremental Nyström method

Abstract

Eigen-decomposition plays a critical role in spectral segmentation. However it is often of low efficiency due to the bottleneck of computation. To solve such bottleneck problem, this paper proposes a dynamic image segmentation method with the incremental Nyström approximation. This method consists of two steps: firstly, we segment the low-resolution images with the original Nyström method; secondly, the high-resolution image newly arriving is dynamically dealt with on the basis of the low-resolution segmentation result. In the second step, an incremental Nyström method is designed to efficiently approximate the eigenvectors of spectral kernel matrices. Experimental results have shown that the proposed method can perform well in image segmentation. © 2011 Springer-Verlag.