Detecting image differences based on reference EMD-filters

Abstract

In our previous works dealing with detecting differences in images in the case of substantial variations in brightness and geometry of an object, we proposed a morphological scheme for image analysis based on diffusion and reference mosaic filters. The filters were defined through the heat kernels of the similarity of image fragments. In this paper, we propose an implementation of this morphological scheme using original reference Earth mover's distance (EMD) filters, in which the optimal matrices of the mutual similarity of mosaic forms are calculated through a linear programming method.