A Robust Face Recognition Framework Based on CapsNet–PCA–ELM Integration

Abstract

In this paper, we propose a hybrid face recognition model based on Capsule Networks (CapsNet) for feature extraction, integrating Principal Component Analysis (PCA) for dimensionality reduction and an Extreme Learning Machine (ELM) as the final classifier. This model aims to address the challenges posed by illumination variations and high-dimensional feature representations, which often degrade the performance of CapsNet when used alone. PCA is employed to reduce noise and improve generalization, while ELM serves as a lightweight and fast classifier. The model was evaluated on a subset of the Yale Face Database containing 15 subjects with 64 images each under different illumination conditions. Experimental results showed that the standalone CapsNet achieved 90.0% accuracy, the CapsNet + ELM model achieved 93.0%, and the proposed CapsNet + PCA + ELM hybrid model achieved the best performance with 96.7% accuracy. These findings demonstrate that combining dimensionality reduction and efficient classification with capsule-based feature extraction provides a more robust and effective solution for face recognition, especially in small or medium-sized datasets.