Robust Keypoint Matching with ORB Features

Basic demo of the papers

[1] Jiayi Ma, Ji Zhao, Jinwen Tian, Alan Yuille, and Zhuowen Tu.<br/>
    Robust Point Matching via Vector Field Consensus,<br/>
    IEEE Transactions on Image Processing, 23(4), pp. 1706-1721, 2014<br/>
    
[2] Jiayi Ma, Ji Zhao, Jinwen Tian, Xiang Bai, and Zhuowen Tu.<br/>
    Regularized Vector Field Learning with Sparse Approximation for Mismatch Removal,<br/>
    Pattern Recognition, 46(12), pp. 3519-3532, 2013<br/>

The original code supplied by this paper uses the SURF detectors which a lot of people have issues compiling due to the copyright issues with the SURF implementation. I adopted their original code to make it working with ORB features.

This also contains usage samples for simple keypoint matching (with Lowe's ratio test and Fundamental-test for outlier rejection). Usage samples are in both C++ and Python. This can be used as a boilerplate code for many computer vision applications. Feature detection and matching is usually very common in a typical computer vision pipeline.

This related github-gist might also be useful.

How to Compile

Only dependency is OpenCV

mkdir build
cmake ..
make
./robust_matcher

Python codes in py directory and can be executed directly.

Result