<< Back

Two-Frame Wide Baseline Matching

This paper describes a novel approach to automatically recover corresponding feature points and epipolar geometry over two wide baseline frames. Our contributions consist of several aspects: First, the use of an affine invariant feature, edge-corner, is introduced to provide a robust and consistent matching primitives. Second, based on SVD decomposition of affine matrix, the affine matching space between two corners can be approximately divided into two independent spaces by rotation angle and scaling factor. Employing this property, a two-stage affine matching algorithm is designed to obtain robust matches over two frames. Third, using the epipolar geometry estimated by these matches, more corresponding feature points are determined. Based on these robust correspondences, the fundamental matrix is refined, and a series of virtual views of the scene are synthesized.

Finally, several experiments are presented to illustrate that a number of robust correspondences can be stably determined for two wide baseline images under significant camera motions with illumination changes, occlusions, and self-similarities. Moreover, using image based rendering technique, a 3D virtual scene is synthesized over these two original images based on the epipolar geometries estimated by proposed method. After testing a number of examples and comparing with the existing methods, the experimental results strongly demonstrate that our matching method outperforms the state-of-art algorithms for all of the test cases.

Associated Publications:

Two-Frame Wide Baseline Matching
The Ninth IEEE International Conference on Computer Vision, Nice, France, 2003.

PowerPoint Presentation


3D case: House, frame 0-6 (60 degree rotation), 150 correspondences,
RMS=0.418 pixels, (from VGG of Oxford University)
[Click for larger image]


Planar case (Homography): Graffiti6, frame 1-5 (60 degree rotation),
202 correspondences, RMS=0.327 pixels  (from INIRA)
[Click for larger image]


Large scaling case: Boat, frame 0-8 (~4 times scaling ),
75 correspondences, RMS=0.686 pixels (from INIRA)
[Click for larger image]


RMS and number of corresponding points of "House'' and "Graffiti-6'' sequences under different viewing angles.


Morphing results using the wide baseline matching (QuickTime format)



More results are available here.