Local Scale Measure for Remote Sensing Images

This paper addresses the problem of defining a scale measure for digital images, that is, the problem of assigning a meaningful scale information to each pixel. We propose a method relying on the set of level lines of an image, the so-called topographic map. We make use of the hierarchical structure of level lines to associate a level line to each pixel, enabling the computation of local scales. This computation is made under the assumption that blur is constant over the image, and therefore adapted to the case of satellite images. We then investigate the link between the proposed definition of local scale and recent methods relying on total variation diffusion. Eventually, we perform various experiments illustrating the spatial accuracy of the proposed approach.