利用vlfeat工具包可行。

返回： pairwise distance matrix D of the columns of S1 and S2, yielding

D = VL_ALLDIST2(X,Y) returns the pairwise distance matrix D of the columns of S1 and S2, yielding

D(i,j) = sum (X(:,i) - Y(:,j)).^2
VL_ALLDIST2(X) returns the pairwise distance matrix fo the columns of S, yielding

D(i,j) = sum (X(:,i) - X(:,j)).^2
VL_ALLDIST2(…,’METRIC’) changes the computed distance. Supported values for METRIC are

METRIC D(i,j)

LINF max |X - Y|
L2 sum (X - Y).^2
L1 sum |X - Y|
L0 sum (X ~= Y)
CHI2 sum (X - Y).^2 ./ (X + Y)
HELL sum (X^.5 - Y^.5) .^ 2
(Notice that the standard definition of chi2 is half of what is computed here).

VL_ALLDIST2(…,’KERNEL’) computes the following ‘kernels’ K:

KERNEL K(i,j)

KL2 sum X .* Y
KL1 sum min (X, Y)
KCHI2 2 * sum (X .* Y) ./ (X + Y)
KHELL (X .* Y) .^ 0.5
The constant are chosen so that D(i,j) = K(i,i) + K(j,j) - 2 K(i,j) where D is the metric corresponding to the kenrel (if the arguments are non-negative vectors). Each kernel can be interpreted as the inner product inducing the corresponding metric in an embedding of the real space into an approrpiate reproducing Kenrel Hilbert space.

VL_ALLDIST2() supports several storage classes. X and Y must have the same storage class. The sotrage class of D is promoted to reduce the chance of overvlow, but this is not checked.