Abstract: In this paper,a novel classifier algorithm based on orthogonal projection was presented. This classifier projecled testing samples to orthogonal subspaces that were spanned by training samples of per class, and calculated the dislance of the testing samples and subspaces,then classified the testing samples according to the distance. The character of this classifier is that it need not calculate the covariance matrices' inverses, so it is very suit for small sample size problem. The experimental results in ORI. database prove this classifier algorithm outperforms the traditional classifier algorithm in recognition rate.

Key words: Orthogonal projection,Classifier,Sma11 sample size problem,Face recognition