Figure-Ground Organization using 3D Symmetry

We present a novel approach to object localization using mirror symmetry as a general purpose and biologically motivated prior. 3D symmetry leads to good segmentation because (i) almost all objects exhibit symmetry, and (ii) configurations of objects are not likely to be symmetric unless they share some additional relationship. Furthermore, psychophysical evidence suggests that the human vision system makes use symmetry in constructing 3D percepts, indicating that symmetry may be important in object localization. No general purpose approach is known for solving 3D symmetry correspondence in 2D camera images, because few invariants exist. Therefore, to test symmetry as a clustering mechanism, we disambiguated the symmetry correspondence problem with the epipolar geometry of the binocular correspondence problem in order to simplify both. Mirror symmetry is a spatially global property that is not likely to be lost in the spatially local noise of binocular depth maps. Furthermore, each of these two correspondence problems provides non-overlapping constraints that makes it easier to solve both at once rather that each individually. We tested our approach on a corpus of 60 images collected indoors with a stereo camera system. K-means clustering was used as a baseline for comparison. The informative nature of the symmetry prior makes it possible to cluster data without a priori knowledge of which objects may appear in the scene, and without knowing how many objects there are in the scene