Adaptive rank-order filters for image processing based on local anisotropy measures

The use of digital image processing techniques for solving application-specific problems confronts the user with a manifold variety of tasks. However, when the image data originate from natural scenes, there are several essential aspects shared by most of these tasks. It is useful to try to recognize these common aspects, because they can serve as guidelines to the development of low-level image processing procedures which can be applied to a wide variety of image data. Although image enhancement and preprocessing for segmentation purposes may appear, from the user’s point of view, as quite different tasks, the difference between them can be considered rather of quantitative than of qualitative nature. In fact, the latter task requires an overenhancement of all the image features that are perceived as discontinuities by the user, for instance, texture transitions, edges, lines, or detail, and an underenhancement of all that is perceived as homogeneous. After a “good” preprocessing for segmentation the image may look quite different from the original one. On the other hand, the enhancement aims in the first place at emphasizing some image patterns or structures for the sake of supporting a visual image interpretation, whereby considerable image alterations are not desired. With the aim of developing low-level adaptive rank-order filters for carrying out in a unified approach the tasks outlined above, the basic requirements to be met by these filters can be outlined as follows: