Template for the function vqlbg.m

function c = vqlbg(d, k)

% VQLBG Vector quantization using the Linde-Buzo-Gray algorithm

% Inputs:

%       d contains training data vectors (one per column)

%       k is number of centroids required

% Outputs:

%       c contains the result VQ codebook (k columns, one for each centroids)

Hints

Cluster Vectors (Nearest-Neighbor Search):

The nearest-neighbor search step is: given a current codebook c, assign each training vector in d with a closest codeword. To do that, one needs to compute the pair-wise distances between each vectors in d to each vectors (codeword) in c. This can be done with the supplied function disteu:

    z = disteu(d, c);

Now z(i, j) would be the distance between the training vector d(:, i) and the codeword c(:, j). Next step, for each training vector, find the closest codeword. For this, use the Matlab function min:

    [m, ind] = min(z, [], 2);

The result index vector ind contains the associated cluster number for each training vector. So to access to all the training vectors that belong to the cluster number j (those vectors that are closest to the codeword c(:, j)), one can use:

    d(:, find(ind == j));

Find Centroids

The centroid of all vectors in a particular cluster is found by the Matlab function mean. For example, after the Nearest-Neighbor Search step above, the new centroids of the clusters number j can be updated as follows:

    c(:, j) = mean(d(:, find(ind == j)), 2);