Study on the performance of moments as invariant descriptors for practical face recognition systems

The performance of pattern recognition systems that use statistical features depends on a specific feature extraction technique. This technique is used to represent an image by a set of features and to reduce the dimension of the image space by removing redundant data. This study investigates a variety of moment-based feature extraction techniques, including Zernike, pseudo Zernike and orthogonal Fourier-Mellin, for the recognition of human faces. In this study, the authors have concerned with both values and orders of moments in the sense of accuracy and efficiency. Two public large face databases, FERET and CAS-PEAL-R1, have been exploited in the experiments. The authors have also employed two typical classifiers, radial basis function neural network and support vector machine, in order to ensure the reliability and consistency of the results from the classification point of view. The extensive experiments in variations of illumination, expression, aging and different accessories have shown that Zernike moments achieve the best overall performance in terms of both classification accuracy and execution time.