A Low-Complexity Deformation Invariant Descriptor

In this paper, we propose a descriptor which is invariant to general deformations (only intensity locations change but not their value) by using Hilbert scanning. In our method, an image is converted to a 1D sequence through Hilbert scanning at first. Then, we embed this sequence as a 1D curve in the 2D space. Because Hilbert scanning preserves the coherence in a 2D image, it is easily to understand that the area under the curve is invariant to intensity location changes, naturally. Hence, we use some areas for an interest point as a deformation invariant descriptor. This descriptor can be computed in the 2D space efficiently than other approaches where an image is embedded in the 3D space or the dimensions of descriptors are very large. The experimental results show that our descriptor is low-complexity and superior to other approaches on interest point matching in deformation images