Tensor correlation filter based class-dependence feature analysis for face recognition

Recently, class-dependence feature analysis (CFA), which is based on the design of correlation filters in the frequency domain, has been developed for robust face recognition. Traditional CFA designs correlation filters by using two-dimensional (2D) Fourier transforms of the images. In this paper, we propose a tensor correlation filter based CFA (TCF-CFA) method to generalize traditional CFA by encoding the image data as tensors. Experimental results on four benchmark face databases show the effectiveness and robustness of TCF-CFA for face recognition.