GMS matching strategy (image)

This sample demonstrates the GMS matching strategy.

Sources:

opencv_contrib/modules/xfeatures2d/samples/gms_matcher.cpp

Options

opts = struct();
opts.nfeatures = 10000;
opts.fastThreshold = 20;
opts.drawSimple = true;
opts.withRotation = false;
opts.withScale = false;

% helper function
siz = @(img) [size(img,2), size(img,1)];

create feature detector and descriptor matcher objects

detector = cv.ORB('MaxFeatures',opts.nfeatures, 'FastThreshold',opts.fastThreshold);
matcher = cv.DescriptorMatcher('BruteForce-Hamming');

pair of images

imgL = imread(fullfile(mexopencv.root(),'test','left01.jpg'));
imgR = imread(fullfile(mexopencv.root(),'test','right01.jpg'));

detect keypoints and compute and match descriptors

[kpRef, descRef] = detector.detectAndCompute(imgL);
[kpCur, descCur] = detector.detectAndCompute(imgR);
matchesAll = matcher.match(descRef, descCur);
fprintf('matches: %d\n', numel(matchesAll));

matches: 5330

GMS matching

tic
matchesGMS = cv.matchGMS(siz(imgL), kpRef, siz(imgR), kpCur, matchesAll, ...
	'WithRotation',opts.withRotation, 'WithScale',opts.withScale);
toc
fprintf('matchesGMS: %d\n', numel(matchesGMS));

Elapsed time is 0.036273 seconds.
matchesGMS: 1044

show results

out = cv.drawMatches(imgL, kpRef, imgR, kpCur, matchesGMS, ...
	'NotDrawSinglePoints',opts.drawSimple);
imshow(out), title('Matches GMS')