Abstract:

Aiming at the unsupervised and nonorthogonal property of the locality preserving projection, a supervised orthogonal locality preserving projection algorithm is proposed. The algorithm considers both of the class information and the orthogonal basis functions. Firstly, the withinclass similarity matrix and the betweenclass similarity matrix are redefined by using the class information, the ratio of the betweenclass scatter and the withinclass scatter is maximized, and the locality structure is preserved effectively. Secondly, the projection vector is orthogonalized, and the space structure of data is reserved as well as the effect of face recognition is further improved. Experimental results on ORL and FETER face databases demonstrate that the effectiveness of our proposed algorithm is better than other algorithms.

Key words: face recognition, image dimensionality reduction, supervised, orthogonalization, locality preserving projection