Matching of images with projective distortion using transform invariant low-rank textures

• A TILT based matching method for images with projective distortion is presented.
• This issue reduces to matching images with similarity distortion by using TILT.
• Geometric shape and local SIFT descriptors are jointly used in the method.
• An automatic low-rank texture region detection method is presented.
• The matching method effectively works for images with projective distortion.

A matching method is presented for images with projective distortion based on transform invariant low-rank textures (TILT). In the method, the problem of matching images with projective distortion is first reduced to a problem of matching two rectified images just with scaling and translation distortions via TILT. Then a point-feature based matching method is employed to establish the corresponding points between the two rectified images. This is different from some traditional methods that try to directly seek local affine or projective invariants from input images. Moreover, no prior knowledge on the epipolar geometry is required in the proposed method. An automatic low-rank texture region detection method is presented to make the method more applicable in practice. Additionally, a new descriptor is constructed by combining a proposed geometric shape descriptor and the traditional SIFT descriptor to further improve the correct matching rate. Experimental results demonstrate the validity of the proposed method.