Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
4629434 | Applied Mathematics and Computation | 2012 | 10 Pages |
Discriminant common vectors (DCV), which can effectively extract the features of face images, is a recently proposed algorithm to overcome the small sample size (SSS) problem encountered by linear discriminant analysis (LDA). Its numerical accuracy is high and computational complexity is low, however, the DCV algorithm is not suitable for online training problems. In order to address this problem, an incremental DCV (IDCV) method is developed in this paper. The IDCV algorithm can incrementally learn the optimal projection matrix instead of recomputing the DCV again when new sample is added into the training set. Theoretical analysis denotes that IDCV is much more efficient that DCV. Experiments on ORL, PIE and AR face databases demonstrate the efficiency of our proposed IDCV algorithm over the original batch DCV algorithm.