Facial expression recognition using Reversible Neural Network

In this article, a concept of Reversible Neural Network, unexplored as yet, is introduced. When identifying facial expressions using a set of salient features such Reversible Neural Network is found to play a crucial role in recognizing the facial expressions into six basic categories. These salient characteristics are extracted by utilizing distance and shape signatures from three benchmark datasets to produce a noteworthy feature set. Such distance and shape signatures are utilized for finding stability indices capable of identifying a particular expression quite effectively. Statistical metrics like skewness, entropy, and moment are computed using distance and shape signatures in order to ensure the discriminative behavior in the selected feature set. The proposed Reversible Neural Network typically receives such discriminative feature set and transforms it into six basic facial expressions. The proposed network is tested and validated on the MMI, CK+, and JAFFE data sets in order to conduct our experiment and confirm the efficacy of facial expression identification. Additionally, its superiority in the context of facial expression recognition is justified by comparison with the state-of-the-art approaches.