Face recognition based on DCT and multi-scale ∊-LBP

A face recognition method based on the discrete cosine transform (DCT) and ϵ-Local Binary Pattern (ϵ-LBP) is presented. The method makes use of DCT to do with the face image and selects the upper-left corner of the DCT matrix as frequency features. At the same time, we adopt to ϵ-LBP extract different scale texture features by using different ϵ, then use ReliefF to reduce its dimensions. Finally, we fuse the feature of the DCT and LBP after normalization, and classify by SVM. Experiments performed on UMIST facial database can achieve an accuracy of 99.15%, results indicate that this method can get a variety of facial feature information to improve the recognition rate by using the complementary facial information of frequency domain and space domain.