Advanced Meta-Analysis of Facial Expression Recognition using Spatio-TS NN Models

A recent increase in the utilization of deep neural networks for learning discriminative representations in automatic Facial Expression Recognition has occurred, propelled by the achievements of deep learning in various fields and the growing intricacy of real-world FER issues. The principal challenges confronting deep FER systems are overfitting stemming from insufficient training data and extraneous variations not associated with facial expressions, such head posture, lighting conditions, or identification bias. A systematic strategy is employed to tackle these issues, according to the sequential order of preprocessing, feature extraction, feature selection, and model training. Preprocessing entails Gaussian convolution and bilateral filtering, with the bilateral filter serving as a weighted average of pixels. Feature extraction employs geometric characteristics obtained from distance measurements between facial landmarks, subsequently applying the adaptive butterfly optimization method for feature selection. The chosen attributes are subsequently utilized to train Spatio-TS NN models for enhanced performance. The proposed methodology markedly surpasses current state-of-theart techniques, with a remarkable accuracy enhancement of 99.15%. This paper introduces a proficient deep learning methodology for facial emotion recognition, systematically tackling critical problems and using optimal preprocessing, feature selection, and Spatio-TS NN training to improve accuracy, rendering it a viable option for practical applications.