GMS (Grid-based Motion Statistics) feature matching strategy

matchesGMS = cv.matchGMS(size1, keypoints1, size2, keypoints2, matches1to2)
matchesGMS = cv.matchGMS(..., 'OptionName', optionValue, ...)

Input

• size1 Size of first image [w,h].
• keypoints1 Keypoints from the first source image. A 1-by-N structure array with the following fields:
  • pt coordinates of the keypoint [x,y]
  • size diameter of the meaningful keypoint neighborhood
  • angle computed orientation of the keypoint (-1 if not applicable). Its possible values are in a range [0,360) degrees. It is measured relative to image coordinate system (y-axis is directed downward), i.e in clockwise.
  • response the response by which the most strong keypoints have been selected. Can be used for further sorting or subsampling.
  • octave octave (pyramid layer) from which the keypoint has been extracted.
  • class_id object id that can be used to clustered keypoints by an object they belong to.
• size2 Size of second image [w,h].
• keypoints2 Keypoints from the second source image. Same format as keypoints1.
• matches1to2 Input 1-nearest neighbor matches from the first image to the second one. A 1-by-M structure array with the following fields:
  • queryIdx query descriptor index (zero-based index)
  • trainIdx train descriptor index (zero-based index)
  • imgIdx train image index (zero-based index)
  • distance distance between descriptors (scalar)

Output

• matchesGMS Matches returned by the GMS matching strategy.

Options

• WithRotation Take rotation transformation into account. default false
• WithScale Take scale transformation into account. default false
• ThresholdFactor The higher, the less matches. default 6.0

GMS feature matching strategy by [Bian2017gms].

Notes

• Since GMS works well when the number of features is large, we recommend to use cv.ORB features and set FastThreshold to 0 to get as many features as possible quickly.
• If matching results are not satisfying, please add more features (we use 10000 for images with 640x480 size).
• If your images have big rotation and scale changes, please set WithRotation or WithScale to true.

References

[Bian2017gms]:

JiaWang Bian, Wen-Yan Lin, Yasuyuki Matsushita, Sai-Kit Yeung, Tan Dat Nguyen, and Ming-Ming Cheng. "GMS: Grid-based motion statistics for fast, ultra-robust feature correspondence". In IEEE Conference on Computer Vision and Pattern Recognition, 2017.