Two-dimensional face recognition algorithms in the frequency domain

Principal component analysis (PCA), well-known for its compaction capability and robustness against noise, is a widely used technique for face recognition. However, it has major drawbacks: (i) losing image details, (ii) having a large time complexity and (iii) suffering from adverse effect of intra-class pose variations. To overcome the last drawback in PCA, Fourier magnitude (FM-PCA) has been proposed in which Fourier magnitudes have been used for feature extraction. Furthermore, to address the other two drawbacks, two-dimensional PCA (2DPCA) algorithms have been proposed. In this paper, to overcome the problems (i) and (ii) in FM-PCA and the problem (iii) in 2DPCA algorithms, Fourier magnitude 2DPCA algorithms which incorporate the advantages of FM-PCA and 2DPCA algorithms are developed. Extensive simulations on the ORL database confirm the effectiveness of the proposed algorithms in providing higher accuracy over PCA, FM-PCA and 2DPCA algorithms with a much smaller complexity compared to that of FM-PCA.