Abstract: Face recognition is an active research area in the artificial intelligence. A face recognition algorithm using improved Non-negative Matrix Factorization(NMF) with Projected Gradient(PG) for single-trial feature extraction is proposed based on this problem. NMF is a matrix factorization method, which can reflect the inherent partial contact and effectively express single sample information. However, NMF iteration time complexity of the gradient projection optimization method significantly reduces the NMF iteration time complexity of the problem. But the single training sample information has inadequate description of disadvantage, for this disadvantage, before the NMF operator, training sample is filtered by multi-orientation Gabor filters with multi-scale to extract their corresponding local Gabor magnitude map, the PGNMF feature of which were constructed to higher dimensional feature vectors. Experimental results on the ORL face database, YALE face database and FERET face database show that the proposed method is feasible and has higher recognition performance compared with GREY, PCA, ICA, NMF, PGNMF and other algorithms where only one sample image per person is available for training.