What distance metric can I use for comparing image features like elongation and solidity of a contour of each image? Except Least Square and without using a support vector machine because i do not know at which class images belong.
$\begingroup$
$\endgroup$
1
-
$\begingroup$ Have you tried using the first two eigenvectors of the contour matrix (or its covariance matrix) as a measure for elongation? $\endgroup$– JunuxxCommented Feb 27, 2013 at 9:59
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
If I understood you correctly, each contour is described by a 2-element feature vector $f = [e, s]$, where $e$ is elongation and $s$ is solidity.
In that case, you might want to try the Mahalanobis distance, which is defined as follows: $$d(f_1, f_2) = \sqrt{(f_2 - f_1)C^{-1}(f_2 - f_1)}$$ where $f_1$ and $f_2$ are the feature vectors that you are comparing, and $C$ is the covariance matrix of your data set.
