Two-dimensional Locality Preserving Projections Image Recognition Algorithm based on Nearest Orthogonal Matrix

Aiming at the recognition problems such as facial changes and uneven grayscale caused by expression and illumination changes in image recognition, a two dimensional Locality Preserving Projections based on Nearest Orthogonal Matrix (2DLPP-NOM) is proposed. 2DLPP-NOM). Firstly, the singular value decomposition is utilized to obtain the nearest orthogonal matrix representation of the image matrix; then the projection matrix is obtained by projecting the nearest orthogonal matrix into the subspace through 2DLPP, and the image features are obtained from this projection. The face image recognition experiments on Yale, AR and CMU-PIE databases demonstrate the significant advantages of this method over the original nearest orthogonal matrix representation.