Grouping objects in multi-band images using an improved eigenvector-based algorithm

Spectral clustering algorithms have attracted considerable attention in recent years. However, a problem still exists. These approaches are too slow to scale to large problem sizes. This paper aims at addressing a coarsening algorithm for efficiently grouping large-dataset objects within multi-band images. The coarsening algorithm is based on random graph theory, and it proceeds by combining local homogeneous resolution cells into a set of irregular blocks so the spectral clustering algorithms run efficiently at some coarse level. For multi-band images, we formulate the similarity between pairwise objects as a novel normalized expression and reformulate it in the form of a matrix so that we can implement our algorithm in a few lines using IDL. Finally, we illustrate two examples in agriculture which confirm the effectiveness and efficiency of the proposed algorithm.