From matching interest points between two images, the algorithm finds a set of reliable correspondences using coherency. Virtual straight lines joining interest points in the same image are encoded by a descriptor invariant w.r.t. certain geometric and photometric deformations. This descriptor should be found in virtual lines between corresponding points in another image for ensuring the coherency of both correspondences.

The algorithm is used as a discrimination step between true/false correspondences in the process of rigid or deformable registration and of stereo reconstruction.