Extraction of significant regions in color images for landmark identification

In this paper, we address the problem of natural landmark characterization in outdoor environments. Our approach assumes that the image has been previously processed in order to detect the most color-salient areas of the image, which are considered as possible candidates to contain a landmark. We take each of these selected areas and perform a color segmentation of them involving only the most relevant regions, which will be used to characterize a possible landmark contained in this area. The re-identification of the same landmarks in successive views should be done in a posterior step by comparing their descriptions, which consist in the color and first and second order moments of each segmented region. The main contribution of this paper is the algorithm for the segmentation of the relevant regions of an image.