Searching of erroneous correspondences of key points

  We implemented two algorithms which searches  erroneous correspondences of key points. 

The 1^st algorithm.

Key points are grouped using the  recursive algorithm breadth-first search. Then for every group calculated average slope angle of the vector points and the average length of a vector. Vectors, whose parameters differ significantly from the average parameters identifies as erroneous.

This algorithm is invariant to rotation and scaling.

Symbols in images:

The 2^nd algorithm.

Pic. 1. Pairs of key points before using the algorithm (yellow rectangles the area that was analyzed on pic. 2).

Pic. 2. Displacement vectors of key points in selected area.

Pic. 3. Pairs of key points after using the second algorithm.