Abstract : Superpixels have become very popular in many computer vision applications. Nevertheless, their use remains under-exploited since the superpixel decomposition may produce irregular and non stable segmentation result due to the dependency of such methods to the image content. In this paper, we first introduce a novel structure, a superpixel-based patch, called SuperPatch. The proposed structure, based on superpixel neighborhood, leads to a robust descriptor since spatial information is naturally included. The generalization of the PatchMatch correspondence method to these SuperPatches, named SuperPatchMatch, is next proposed. Image segmentation and labeling experiments show the potential of our approach, since we obtain very competitive results compared to the state-of-the-art methods.