This is code provides a minimal working (MWE) example of the paper "Using Spatial Order to Boost the Elimination of Incorrect Feature Matches" by Lior Talker, Yael Moses and Ilan Shimshoni, presented at CVPR 2016. For questions or information about the paper or code please contact Lior Talker (liortalker@gmail.com or https://liortalker.wix.com/liortalker) Only for academic or other non-commercial purposes (under GPL terms).

Running the algorithm:

For the synthetic experiments run SyntheticExp1.m or SyntheticExp2.m (Provide only the number of matches and (for Exp1 the number of inliers)

For a real experiments run either runKendallInlierRate.m (no parameters - loads im1.jpg and im2.jpg), or runKendallInlierRateOnTwoImages.m for which the input is two images and parameters, or estimateKendallInlierRateSeparateWindows.m (or estimateKendallInlierRateJointlyWindows.m) for which the input is two sequences and a resolution parameter usually = 0.1)

Dependencies:

The algorithm itself has no dependencies, but to extract features (e.g., SIFT) the example is dependent on VLFeat. You can download it from https://www.vlfeat.org, place it in a folder inside the project, invoke vl_setup, and run the example itself.