On the impact of outliers on high-dimensional data analysis methods for face recognition

In this paper, the impact of outliers on the performance of high-dimensional data analysis methods is studied in the context of face recognition. Most of the existing face recognition methods are based on PCA-like methods: Faces are projected into a lower dimensional space in which similarity between faces is more easily evaluated. These methods are, however, very sensitive to the quality of face images used in the training and the recognition phases. Their performance significantly degrades when faces are not well centered or taken under variable illumination conditions. In this paper, we study this phenomenon for two face recognition methods (PCA and LDA2D) and we propose a filtering process that allows an automatic isolation of noisy faces which are responsible for the performance degradation. This process is performed during the training phase as well as the recognition phase. It is based-on the recently proposed robust high-dimensional data analysis method RobPCA. Experiments show that this filtering process improves the recognition rate by 10 to 20%.