How to implement the Nearest Subspace algorithm

Dear all, I am comparing the performance of different classifiers on the same train and test samples: the Nearest Neightbour (NN), Nearest Subspace (NS) and Support Vector Machine (SVM).

I implemented the NS simply # by projecting the test sample into the subpace # comparing the norms, and # assigning the test sample to the nearby subspace.

In pseudo-code:

winner_subspace=1;
proj_test=test'*subspace_1
for i=2: total_number_subspace
    Calculate the subspace_i
    proj_temp=test'*subspace_i
    if norm(proj_temp)<norm(proj_test)
       winner_subspace=i;
    end
end

The obtained performance is too low ... where is the mistake? Thank you in advance, Idil