Optimized face-emotion learning using convolutional neural network and binary whale optimization

Human emotion detection using facial expressions might be easy for humans, but computing technology to accomplish the same task is more challenging. We can recognize emotions from images using the latest computer vision and machine learning (ML) advancements. This research proposes a novel optimized face emotion learning method with binary whale optimization (OFELBW). The OFELBW is implemented in three phases, the first phase with a convolutional neural network (CNN) in which from the image the background noise is removed in the initial phase, and the facial feature extraction is performed in the second phase. Finally, the binary whale optimization algorithm is used for the feature selection to obtain the most relevant feature subset. The proposed OFELBW method was examined with more than 750 K images using SFEW, CK+, JAFFE, and FERG datasets. We have compared our proposed OFELBW model with other existing techniques to examine the accuracy of our models with the above-mentioned datasets and received an accuracy of 98.35% with the CK+ dataset, 99.42% with the FERG dataset, 96.6% with the JAFFE dataset and 64.98% with the SFEW with 80% training, 10% testing, and 10% validation set. This technique will be useful in various applications such as human social/physiological interaction systems, mental disease diagnosis and military environment, etc.