Minimum spanning tree adaptive image filtering

The main focus of this paper is related to anisotropic morphological edge preserving filters. We present in this work neighborhood filters defined on the minimal spanning tree (MST) of an image (according to a local dissimilarity measure between adjacent pixels). The designed filters take advantage of the property of the MST to detect and follow the local features of an image. This approach leads to neighborhood filters where the structuring elements adapt their shape to the minimal spanning tree structure and therefore to the local image features. We demonstrate the quality of this method on natural and synthetic images.