Feature matching

Source: http://medium.com/data-breach/introduction-to-feature-detection-and-matching-65e27179885d Backlinks:  Bag of words , sparse/feature-based-vslam

For matching between images, i.e. to establish a relationship (‘correspondence’) between two images of the same scene or object.

Basic algorithm

• Find/detect a set of identifying (‘distinctive’) keypoints from all images to be matched
• Define a search region around each keypoint
• Extract and normalise the region content
• Compute a local descriptor from the normalised region
• Match local descriptors between the images

Performance of matching methods depend on

• characteristic of the interest points themselves
• choice of the feature detectors / image descriptors
  • e.g. corner detector for man-made surfaces, blob detector for more organic sturctures
  • the detector/descriptor should be able to handle image degradation

Some algorithms

• Brute-Force Matcher
• FLANN (Fast Lirbary for Approximate Nearest Neighbours) Matcher