In this paper, cross-weighted centroid (CWC) is constructed to extract affine invariant features. Every point in an image is assigned with a cross-weight based on the distribution of the image. CWC, derived from these weighted points, is affine invariance. Based on the original centroid and CWC of the image, a series of centroids of some affine regions can be obtained by iterating affine region cutting, i.e., ARC. Several affine invariant triangles are constructed. Consequently, affine invariant features can be derived by the area of these triangles. Experiment results show the efficiency of the proposed method.