Community Detection for Hierarchical Image Segmentation

In this paper, we present a new graph-based technique to detect segments or contours of objects in a given picture. Our algorithm is designed as an approximation of the Louvain method that unfolds the community structures in a large graph. Without any a priori knowledge on the input picture, relevant regions are extracted while the optimal definition of a contour, depending on the user or the application, can be tuned using parameters. The communities found are also hierarchical allowing to find subregions inside an object. We present experimental results of our method on real images.