A General Preprocessing Method for Improved Performance of Epipolar Geometry Estimation Algorithms

In this paper a deterministic preprocessing algorithm is presented, whose output can be given as input to most state-of-the-art epipolar geometry estimation algorithms, improving their results considerably. They are now able to succeed on hard cases for which they failed before. The algorithm consists of three steps, whose scope changes from local to global. In the local step it extracts from a pair of images local features (e.g. SIFT). Similar features from each image are clustered and the clusters are matched yielding a large number of putative matches. In the second step pairs of spatially close features (called 2keypoints) are matched and ranked by a classifier. The 2keypoint matches with the highest ranks are selected. In the global step, from each two 2keypoint matches a fundamental matrix is computed. As quite a few of the matrices are generated from correct matches they are used to rank the putative matches found in the first step. For each match the number of fundamental matrices, for which it approximately satisfies the epipolar constraint, is calculated. This set of matches is combined with the putative matches generated by standard methods and their probabilities to be correct are estimated by a classifier. These are then given as input to state-of-the-art epipolar geometry estimation algorithms such as BEEM, BLOGS and USAC yielding much better results than the original algorithms. This was shown in extensive testing performed on almost 900 image pairs from six publicly available data-sets.