Utilizing the graph modularity to blind cluster multispectral satellite imagery

The fully automatic separation of spectral image data into clusters is a problem with a wide variety of desired and potential solutions. Contrary to the typical approaches of utilizing the first order statistics and Gaussian modeling of spectral image data, the method described in this paper utilizes the spectral data structure to generate a graph representation of the image and then clusters the data by applying the method of optimal modularity for finding communities within the graph. After defining and identifying pixel adjacencies to represent an image as an adjacency matrix, a quantity known as the graph modularity is maximized to split the data into groups of spectrally similar pixels. Recursion with the subgroups created in each split creates the data clustering. The groups where the maximal modularity is not above a given threshold are not split, resulting in a stopping condition and an estimation of the number of clusters necessary to fully describe the data. By ignoring any reliance on assumptions of the shape of the data, this method excels in regions where typical clustering methods fail, such as high resolution urban scenes with very high clutter and regions with subtle variability like coastal bodies of water.