Figure 1: Scale-space matching extends gradient support. Given an image (a) of a disk we recover its position θ on the horizontal axis. At stationary resolution (σ = 0), the initial and target (dotted) disks do not overlap, as shown in the corresponding 1D signals in (b). The image gradient ∂I/∂θ is sparse (orange) and is non-zero only at the boundaries of the disk (c-top). The error gradient ∂E/∂θ is zero everywhere (green) and the optimization is stuck in a local minimum. When matching at coarser scales (d), the gradients are no longer sparse (c-bottom), leading to optimal recovery.