Enhancing Facial Emotion Recognition Using DCNN through Effective Extraction for High-Level Features

Abstract

Abstract Views: 0

Facial Emotion Recognition, or FER, is increasingly important in human-computer communication, psychology, and advanced monitoring. Researchers seek to enhance accuracy of recognition with high-level feature extraction techniques, utilizing the advancements made in deep convolutional neural networks (DCNNs). This systematic literature review (SLR) compiles and analyses forty-two (42) studies published from 2020 to 2025 across IEEE Xplore, SpringerLink, ScienceDirect, Scopus, and ACM Digital Library. The review marks trends in DCNN-based FER and analyses the datasets used within the research such as FER-2013, CK+, JAFFE, along with their respective metrics and feature extraction methods. The DCNNs were quantitatively found to outperform other architectures, although interpretable and deployable systems were lacking. Through qualitative synthesis, the major focus was placed on feature engineering with deep learning, merging models, and creating comprehensive pre-processing workflows. The review indicates that while steps have been taken toward achieving the broad-scope goals, the diversity of datasets, unified metrics, and real-time systems hinder progress. The comparison of the approaches for high-level feature extraction builds the foundation of the review, making it distinct.  This SLR aims to guide future efforts by highlighting unresolved issues and recommending directions for practical and responsible FER systems.