Performance Evaluation Of Subspace-based Face Recognition

As a typical biometric technology, face recognition has a wide range of applications in many areas, such as business, surveillance, video retrieval, and so on. Compared with other biometrics, face recognition has direct,friendly and convenient advantages in data collection, so people pay more attention to it. Among the traditional recognition techniques, the subspace-based algorithms are well made use of and have good performances.In this paper, we introduce six recognition algorithms, which are based on principal component analysis (PCA), linear discriminate analysis (LDA), Nonnegative Matrix Factor (NMF) and their perspective improved algorithms, L1-PCA, R-LDA and LNMF. On different face database, these algorithms are tested. Experiments indicate that the improved algorithms are all better than the traditional ones. And the performance of linear discriminate analysis is better than others, with good robustness.For the stability evaluation of algorithms, the subset bootstrap algorithm, an improved bootstrap, is suitable for face samples which don't meet the condition of independent identically distributed. With the confidence interval of equal error rate being the stability index, we compare the subset bootstrap technique with the conventional methods. Experiment results show that the subset bootstrap method gives accurate evaluation of stability performance. The face recognition based on R-LDA shows the best robustness with the smallest EER, but the best stable one is traditional LDA.