Fig.shows an example of region-sensitive feature selection for image registration. Fig. 2(a)-(b) shows the detected subregions with dense textures (in green). Fig. 2(c) shows the originally detected local features in a photo. Note that many of the strong features are in the trees/leaves areas. As shown in Fig. 2(d). After feature selection, the remaining features correspond to geometrically important structures in the photos, which result in more robust feature registration as well as faster convergence of the RANSAC algorithm.