Mean shift based segmentation of full VHR imagery with limited resources: An exact solution

This paper adresses the problem of piecewise computation of a segmentation algorithm for large images, without creating any artifacts in the results. We first define a property of segmentation algorithms called stability, and show that the well-known MeanShift algorithm can be derived to be stable. We then propose a segmentation methodology that allows for piecewise segmentation of large images with a stable segmentation algorithm, with the guarantee of identical results with respect to full processing at once. Finally, we introduce two examples of applications: large-scale object based classification, and large-scale object counting.