Large Margin Graph Embedding-Based Discriminant Dimensionality Reduction

Algorithm 1

LMGE-DDR.

	Input: a training set , and the dimensionality of discriminant subspace r.
Output: projection matrix P;
(1)	Construct the intraclass adjacency graph by:
	and interclass adjacency graph by:
(2)	computing the based on (15)
(3)	Eigendecompose the matrix, where
(4)	Choose the r largest eigenvalues corresponding eigenvectors: .