Inlier selection, i.e., the extraction of small set of correct fiduciary correspondences from a large set of points is a fundamental step of various estimation processes in Computer Vision, ranging from object recognition, to surface/image registration, to pose estimation. The typical approach is to attach to each point a descriptor which is used in aiding the extraction of good correspondences. However, unary descriptors alone cannot guarantee the lack of outliers, which must then be filtered out. Typically these filtering approaches are either based on the initial estimation performed with the outliers, or on a RANSAC-like process, thus being effective only for small outlier ratios. We offer a change in perspective, where a game-theoretic matching technique that exploits global geometric consistency allows to obtain an extremely robust correspondences even when coupled with simple features exhibiting very low distinctiveness. The effectiveness of the approach is shown on surface registration and pose estimation problems.