Computing Eigenface from Edge Images for Face Recognition Based on Hausdorff Distance

The different face regions have different degrees of importance for face recognition. In previous Hausdorff distance measures for face recognition, such as Spatially Eigen-Weighted Hausdorff distance (SEWHD) in which the weighting function is computed from grayscale images. However, Hausdorff distance is a measure for two binary point sets, not for grayscale point sets. Based on our previous work on Weighted Hausdorff distance (EFWHD) for face localization, a new weighting function of Hausdorff distance measure termed Edge Eigenface Weighted Hausdorff distance (EEWHD) is proposed for face recognition in this paper. The weighting function, which reflects the discriminative properties of face edge images effectively, is based on the eigenface of face edge images, not the eigenface of grayscale images in SEWHD, nor the edge points appearing frequency in EFWHD. The weighted Hausdorff distance Experimental results show the new method achieves higher recognition rate comparing with previous Hausdorff distance measures.