brede_mat_onmf - Orthogonal non-negative matrix factorization
[W,H] = brede_mat_onmf(M)
Input: M 'Mat' structure or numerical matrix
Property: Components [ round(sqrt(min(size(X)))) | A positive
integer
Output: W 'Mat' structure
H 'Mat' structure
Orthogonal non-negative matrix factorization. The columns of W
will be orthogonal, likewise for the rows of H, i.e., there will
at most be one positive value in each row of W, and at most one
positive value in each column of H.
A simple algorithm with a winner-take-all function is presently
used, and this approach may not produce optimal results.
Example:
X = randn(12,10);
[W,H] = brede_mat_onmf(X);
See also BREDE, BREDE_MAT, BREDE_MAT_NMF, BREDE_MAT_WTA.
$Id: brede_mat_onmf.m,v 1.2 2008/06/17 16:16:34 fn Exp $