Approach on Sparse Samples Face Retrieval

Face retrieval training samples is difficult to obtain in many situations. Therefore, an effective way to describe the human face becomes more important. We have fused Gabor feature, LGBP feature and LPQ feature to describe face in this paper. Faces are divided into blocks according to features, than fusion features obtained to characterize each individual face. Taken together, our results suggest that fused features for data retrieval have a certain improvement compared with the current popular features in FERET, AT&T database and ours dataset.