Face recognition by subspace analysis of 2D Log-Gabor wavelets features

In this paper, we discuss a face recognition scheme by subspace analysis of 2D log-Gabor wavelets features. In which, an input face image is firstly decomposed with a set of two dimensional log-Gabor wavelets (2D-LGWs) localized with respect to spatial location, orientation and frequency. Based on complex responses of filters, local energy model (LEM) is used to represent log-Gabor features (LGFs) which are substantially effective for the task of recognition. Then, subspace modeling is performed to transform the high dimensional LGFs into more compact one to simplify the task of classification. Common nearest-neighbor (NN) based matching algorithm is adopted to classify a probe to one of classes. The superiority of the proposed scheme for face recognition is comparatively demonstrated with the traditional appearance-based methods. Moreover, performances of several leading subspace techniques, PCA, ICA and LDA, are comparatively evaluated based on LGFs representation.