Facial Expression Recognition in the Wild using Artificial Rabbits Optimizer based Residual Neural Network

• Published in: 2023 International Conference on Advances in Computation, Communication and Information Technology (ICAICCIT) pp. 488 - 493
• Main Authors: Metkewar, P S, Uma, N, Dhanaraj, Rajesh Kumar, Prabhu Kavin, Balasubramanian, Sathyamoorthy, Malathy
• Format: Conference Proceeding
• Language: English
• Published: IEEE 23.11.2023
• Subjects: Adaptation models; Adaptive equalizers; Artificial Rabbits Optimization; Computational modeling; Contrast Limited Adaptive Histogram Equalization; Emotion recognition; Face recognition; Histograms; Rabbits; Residual Neural Network; Stochastic gradient descent; Training
• DOI: 10.1109/ICAICCIT60255.2023.10466005

In affective computing, emotion acknowledgement in the wild is a much-studied area. Although there have been advancements, the difficulty of emotion acknowledgement in the wild due to head movement, face deformation, illumination fluctuation, etc. remains an open subject. To improve the model's simplification ability and, by extension, its performance across a variety of learning tasks, it is crucial that a wide variety of features be extracted by the deep neural network. Interest in facial expression detection in natural settings has grown in recent years despite the difficulty of obtaining discriminative and informative characteristics from partially obscured photos. In the first stage of this paper's pre-processing, the noisy pictures are enhanced by Contrast Limited Adaptive Histogram Equalization (CLAHE). After that, Residual Neural Network (ResNet) features extraction is used, and the same method serves as an emotion classifier. An iterative technique called stochastic gradient descent (SGD) is utilized to refine the ResNet model's objective function. Finally, the Artificial Rabbits Optimization Algorithm (ARO) is used to choose the best value for, thereby enhancing the reliability of the categorization. The suggested method is effective, as evidenced by experimental findings on three popular facial expression recognition in-the-wild datasets: AffectNet, AFEW Dataset, and RAF-DB, where it accomplishes state-of-the-art presentation with 96% accuracy and improves upon existing models by roughly 5% to 8%.