An Optimal Hybrid Deep Learning-Aided Facial Emotion Detection and Classification Scheme to Identify Criminal Activities

Abstract In general, the most significant field of research presently is identification & recognition of facial expressions or emotions. Moreover, recognition & categorization of face emotion are vital in several areas of research like criminal activities investigation, innovative card application, security, surveillance system, and so on. Among these, criminal investigation plays a vibrant part. Since there exists several methods on facial emotion/expression recognition (FER) system, however there were some drawbacks like low prediction rate, lower recognition rate, high error rate and so on. For rectifying these existing issues, a new enhanced optimal DL based model is presented in this manuscript. In this work, input facial dataset is extracted and are preprocessed using Weighted fuzzy Histogram Equalization (WF-HE). From this, the features are extracted using Deep CNN followed by Enhanced glowworm swarm optimization (EGSO)-based feature selection model at which hyper-parameter tuning is carried by attaining fitness function values. This in turn enhances the performance of classifier. The categorization for FER system is carried using Hybrid Deep Variational LSTM (DVLSTM) and DenseNet model. The results are estimated in terms of various performance measures like precision, Area under Curve (AUC), accuracy, F-Measure, sensitivity, specificity and recall, PPV, and error rate. The analysis is made on three input datasets like JAFFE, Extended CK+, and FER2013 dataset. The comparison for attained outcome is made with traditional models to validate proposed system efficiency over other compared schemes.