Design And Verification Of A Face Recognition Algorithm On Small Sample Set

Face recognition technology has become one of the most attractivebiometrics-based technologies and achieved considerable development in recentyears. Various methods for face recognition have been proposed. However, there arestill many thorny issues to be solved in practical applications for face recognitionmethods. The small sample size problem is a typical issue.The small sample size problem implies the number of training samples is sofew that most face recognition methods cannot play its desired performance.Typically, more training samples are helpful to achieve higher recognition accuracy.However, the amount of training samples is usually limited because of variousfactors in practical applications. Based on the analysis of previous research work, anovel virtual samples-based sparse representation method for face recognition(VSFR) has been proposed to deal with the small sample size problem. Theproposed method contains two steps: in the first step, virtual samples are constructedby adding random noise on original training samples, which enlarge the originaltraining set; in the second step, the sparse representation classification method isapplied to face recognition. A large amount of experiments have demonstrated thatVSFR works better on small training samples size sets, and outperforms over manyrespective representation-based methods for face recognition.What’s more, a method also has been proposed to construct partial-occlusionvirtual samples, and is used to improve the performance of VSFR method onpartial-occlusion samples set. Experiments on the AR database have shown thatimproved VSFR outperforms over original VSFR on small samples set withpartial-occlusion. The recognition performance and applicability of VSFR methodhas been further improved.