Chaotic Archerfish Optimizer–Enhanced Double‐Attention Dense BiGRU for Video‐Based Facial Expression Recognition

Facial expression recognition (FER) is a growing and important area of pattern recognition research. In today’s application domains, such as the identification of human facial expression, deep learning techniques have gained popularity. The expressions on people’s faces reveal their motives and emotional actions. An expression is the easiest and most efficient form of nonverbal communication. The accuracy of facial expression detection techniques decreases when faces are partially obscured or the illumination is uneven. To address the issues in the existing systems, a novel optimal hybrid deep learning method is presented in this paper. Here, chaotic archerfish hunting optimizer–based double attention dense bidirectional gated recurrent unit (ChAO‐based DAttn_DenseBi_Net) is introduced to detect the FER from videos. The proposed method is designed by integrating the double attention mechanism based on spatial and temporal attentions with the DenseNet and BiGRU, which is tuned using the proposed ChAO algorithm to enhance the recognition accuracy. The proposed ChAO‐based DAttn_DenseBi_Net technique for FER in three different datasets: RAVDESS, iCVMEFED, and AFEW. With 90% training data, the proposed ChAO‐based DAttn_DenseBi_Net approach attains accuracies of 95.23% for RAVDESS, 96.21% for iCVMEFED, and 95.38% for AFEW. The performance of the proposed method is measured based on various assessment measures such as accuracy, specificity, precision, recall, F1‐measure, and MSE and acquired the values of 96.21%, 97.36%, 94.93%, 96.21%, 95.55%, and 3.79 respectively.