Face recognition analysis for noise images based on combinational mirror-like odd and even features

Since mirror-like odd and even features in face recognition reflect the symmetrical and asymmetrical image information, respectively, their proper combination can improve the recognition rates to some extent. However, the face imaging process can easily be affected by external factors and encounter the noise signal, which disturbs the effect of face recognition based on combinational mirror-like odd and even image features. This paper studies the appropriate combination of features extracted from the mirror-like odd and even face images under the noise condition, respectively, with kernel principal component analysis (KPCA) method. The recognition effects of different combinational methods are studied. Experiments show that when the noise signals are added into the face images, the composition of the features from the mirror-like odd images and the mirror-like even images will change in order to keep good recognition effect, that is, the proportion of mirror-like even features should be increased in the combinational eigenvectors, compared with the mirror-like odd features. However, due to mirror-like even symmetry images not focusing on expressing the image details, their proportion should not be too large.