Similariy matrics in SIFT or SURF

In LOCAL BINARY PATTERNS, the extracted vector have fixed row and col (1x59) so it is easy to compute the distance between two images using Euclidean. However,In SIFT the points are vary image to image. How can calculate the distance or dis(similarity) between Sift Features?

Answer 1

[Moksh]

Hi Awais,

I understand while extracting LBP features, vectors have fixed dimensions so Euclidean distance can be computed. But while extracting SURF features, the SURF points may vary from image to image which may cause the dimensions of features to vary.

You can use the “matchFeatures” function in MATLAB, which returns the indices of matching features for the two input feature sets. This function supports binary features as well as matrices as features.

Below is an example code for finding similar SURF features for two images:

% Defining two images
I1 = imread("cameraman.tif");
I2 = imresize(imrotate(I1, -20), 1.2);  % Rotated cameraman image


% Extracting SURF features for both images
p1 = detectSURFFeatures(I1);
p2 = detectSURFFeatures(I2);
[f1, v1] = extractFeatures(I1, p1);
[f2, v2] = extractFeatures(I2, p2);


% Extracting indexes of matched features
indexPairs = matchFeatures(f1, f2);

MatchedPtsI1 = v1(indexPairs(:, 1));   % Matching points on I1
MatchedPtsI2 = v2(indexPairs(:, 2));   % Matching points on I2

Please refer to the below documentation for more information about the “matchFeatures”, “detectSURFFeatures” and “extractFeatures” functions respectively:

• https://in.mathworks.com/help/vision/ref/matchfeatures.html
• https://in.mathworks.com/help/vision/ref/detectsurffeatures.html
• https://in.mathworks.com/help/vision/ref/extractfeatures.html

Hope this helps!

Best Regards,

Moksh Aggarwal