Image KeyPoint Matching

Please adhere to your organization's rules for using this code. For Concordia University, these rules can be found here.

Feature Detection by Harris corner detector algorithm (Wikipedia)

• Color to grayscale
• Spatial derivative calculation
• Structure tensor setup
• Harris response calculation
• Non-maximum suppression

=> we display the keypoints using a custom drawKeypoints as the opencv version i installed has binding issue on that method

Feature Description

• Compute gradient matrix of the image (dx, dy, magnitude, angle)
• Iterate every key point of the image
  • We get the 16x16 window of the key point
  • we compute the rotation of the patch
  • key_point_descriptions = compute_histogram(array of 4 sub patches of the 16x16 patch) (Each sub patch is 4x4 patch of 16 pixel, each histogram is 8 element array. (so total 16 pixels x 8 elements = 128)

Feature Matching

• Iterate every keypoint with its descriptor

  • Compute SSD of key point with every other key point in other image
  • Find 1st and 2nd Best Key point
  • If the 1st best SSD / 2nd best SSD ratio is < 0.5 and best SSD is < 0.3, We can add to matches list

• We sort this list with respect to there distance

=> display the best 25 matches using drawMatches function