Facial Expression Recognition from Low Resolution Facial Segments Using Pre-trained Networks

Abstract

Understanding the different categories of facial expressions is crucial to comprehend human cognition as well as for designing these computational interfaces. Deep learning is an exciting area in machine learning and has seen an exponential rise in recent years in various applications. In this paper, we use deep learning networks to identify facial expressions in low resolution images. The major contribution of this paper is in analysing the effect of spatial resolution reduction on the facial expression recognition rates. There is not much work reported in this area. Detailed analysis to establish a relationship between recognition rates and drop in spatial resolution is carried out. We have successfully demonstrated that even with extreme spatial resolution reduction in full face images accurate facial expression recognition can be achieved using the proposed frameworks, and there is negligible effect on recognition rates. In addition to full face images, we have also successfully demonstrated that the proposed networks work well for various segments of the face. The results obtained stay consistent even for extreme reduction in spatial resolution of these facial segments. In order to evaluate performance of the proposed frameworks, rigorous sets of experiments are presented in the paper. We can conclude that even when human computer interface systems have access to entire fontal face images, only a small segment of the image is required for accurate facial expression recognition, and the results obtained are consistent even under extreme spatial resolution reductions.