Fusion of Log-Gabor wavelet and orthogonal locality sensitive discriminant analysis for face recognition

Local Sensitive Discriminant Analysis (LSDA) algorithm aims at finding a projection by maximizing the margin between data points from different classes in each local area. However, a major disadvantage of LSDA is that LSDA is non-orthogonal and this makes it difficult to estimate the intrinsic dimensionality and reconstruct the face data. Non-orthogonality distorts the local geometrical structure of the data submanifold. In this paper, an Orthogonal LSDA algorithm is proposed to preserve the local geometrical structure by computing the mutually orthogonal basis functions iteratively. First, the Log-Gabor wavelet is used to extract their corresponding Log-Gabor magnitude features (LGMF) by convolving the normalized face image with multi-scale and multi-orientation Log-Gabor filters. Then, OLSDA operates on LGMFs to extract the discriminative submanifolds. Furthermore, the nearest distance classifier is adopted for classification. Experiments based on the ORL and Yale face databases show the impressive performance of the proposed method.