Adaptive Face Recognition and Emotion Analysis Framework (AFREAF): A Real-Time Multi-Modal Deep Learning System for Comprehensive Facial Attribute Detection

Abstract

This paper presents the Adaptive Face Recognition and Emotion Analysis Framework (AFREAF), a comprehensive, real-time system for simultaneously detecting and analysing multiple facial attributes, including age, gender, and emotional states. AFREAF integrates advanced deep neural networks with adaptive preprocessing techniques, including face alignment via MediaPipe landmarks and histogram equalisation to enhance feature extraction. The framework employs OpenCV's DNN module for robust face detection, specialised convolutional neural networks for age and gender classification, and the FER+ model for emotion recognition. A novel temporal smoothing algorithm ensures prediction stability across video frames. Experimental evaluation demonstrates that AFREAF achieves real-time performance at 28.4 FPS while maintaining competitive accuracy rates of 68.5% for age estimation, 94.2% for gender classification, and 71.3% for emotion recognition. The modular architecture facilitates easy integration into diverse applications, including human-computer interaction, security systems, and behavioural analytics.