A Performance Comparison of the ZM, PZM and LM in the Face Recognition System in Presence of Salt-pepper Noise

This paper compares performance of the feature extraction techniques, Zernike moment (ZM), pseudo-Zernike moment (PZM) and Legendre moment (LM), in the application of face recognition and in presence of salt-pepper noise. In this study, after preprocessing and face localization of an image, its features are extracted. Also RBF neural network (RBFNN) with HLA learning algorithm has been used as a classifier. We trained the classifier three times for each group of extracted features of images. Then we added salt-pepper noises to images with three different probabilities, 0.02, 0.05 and 0.08. The trained RBFNN is tested with original and noisy versions of images. Experimental results on AUTDB show that the performance of the LM in all cases is better than the others.