Chandan Singh, Ekta Walia, and Neerja Mittal
Face Recognition, Zernike Moments, Pseudo Zernike Moments, Euclidean distance, Optimal Similarity Measure
Zernike moments (ZMs) and Pseudo Zernike moments (PZMs) are one of the most frequently used global feature extraction methods in many pattern recognition and image analysis applications. They posses the useful characteristics relating to image description capability, dimension reduction, invariance to image rotation and noise etc. In past, only magnitude coefficients of these approaches are used as rotation invariant image features and their phase coefficients are ignored which results into loss of useful information for classification. In addition to the requirement of invariant and discriminative features for classification, choosing an appropriate distance metric is also an essential step. This study investigates the performance of ZMs and PZMs descriptors using the optimal similarity measure in comparison to the performance of classical Euclidean distance measure (L2 - norm ) for classification of the face images. The optimal similarity measure in itself uses both magnitude and phase coefficients of these descriptors. This differs from the traditional approaches that only compare the distance between magnitude features using L2 - norm. Above all, the robustness of these approaches is analyzed for pose, illumination, expression and noise variations in application to the face recognition system. The optimal similarity measure generates superior results over the L2 - norm metric especially in case of expression and illumination variation.
Important Links:
Go Back
