Learning-Deep-Learning

Classical Keypoints

May 2020

tl;dr: A summary of classical keypoints and descriptors.

BRIEF

• BRIEF: Binary Robust Independent Elementary Features ECCV 2010
• The sampling pattern is randomly generated but the same for all image patches. In openCV, the sampling sequence is pre-fixed.
  • patch window is 31 x 31.
• blog review on BRIEF descriptor

ORB

• ORB: an efficient alternative to SIFT or SURF ICCV 2011
• Sampling pairs should have uncorrelation and high variance to ensure the fixed length would encode maximum discriminative information.
• ORB is improved BRIEF:
  • ORB uses an orientation compensation mechanism, making it rotation invariant.
  • ORB learns the optimal sampling pairs, whereas BRIEF uses randomly chosen sampling pairs.
• blog review on ORB

Key ideas

• Summaries of the key ideas

References

• nterest Point Detector and Feature Descriptor Survey