Investigating the Effect of Noise on Facial Expression Recognition

In this paper we investigate the effect of noise on automated recognition of facial expressions. We take images from a publicly available data set; corrupt them with different noise levels and use them for training and testing algorithms for expression recognition. We do recognition using a variant of AlexNet. We do training and testing on same noise levels and also, train on clean images and test on images with added noise. We show that the recognition performance is fairly robust for reasonable levels of noise, however it degrades considerably after that.