Enhancing Facial Recognition and Expression Analysis With Unified Zero-Shot and Deep Learning Techniques

This paper presents a new model that enhances face recognition while addressing challenges such as illumination variation and noise in facial images. The model integrates state-of-the-art illumination normalization based on Retinex theory with median filtering to improve noise reduction and recognition accuracy. Utilizing pre-trained VGG-16 weights, it incorporates aspects of the Boltzmann machine to extract intricate facial features while reducing overfitting. Additionally, this paper merges zero-shot facial expression recognition methodologies with illumination-robust face recognition within a unified facial analysis framework. Integrating pre-processing models with large language models and advanced feature extraction techniques enables the model to excel across various benchmarks, including FERET, LFW, CelebA, and VGGFace2 datasets. These findings indicate that combining superior normalization techniques with zero-shot learning significantly enhances face recognition and expression analysis under challenging real-world conditions. This solution is practical and robust against variations in illumination and noise encountered in everyday situations.