Robust Symmetry Detection in Natural Images

We propose a perceptually plausible mechanism for symmetry detection in natural images that consists of two steps: 1) A fast attentional step that creates a symmetry attention map followed by 2) a robust symmetry detection step that uses image edge statistics. The input is an image edge map, Ie, and the final output is a set of symmetry axes A = {a1, . . . , aS} with ai = ((xi, yi), oi) (centroid location and orientation) for the ith symmetry axis.