General Information

Graph-partition based algorithms are widely used for image segmentation. We propose an improved graph-partition segmentation method based on a key notion from complex network analysis: partition modularity. In particular, we show how optimizing the modularity measure can automatically determine the number of segments as well as their respective structure—greatly reducing the level of human intervention in the image segmentation process. We furthermore develop an efficient spectral approach that allows for a fast segmentation procedure. The proposed method is simple, efficient, and provides a practical tool for analyzing real-world images.