Hybrid Reinforcement Learning With Optimized SARSA for Improved Face Recognition Systems

Face recognition is a key technique in modern image processing, yet it faces challenges such as achieving high accuracy, reducing computational time, and optimizing memory usage. This research proposes a hybrid model that integrates an enhanced State‐Action‐Reward‐State‐Action (SARSA) reinforcement learning (RL) framework to address these challenges in face recognition tasks. The model utilizes principal component analysis (PCA) for dimensionality reduction and initial feature extraction, followed by a SARSA‐based online Q‐learning algorithm to refine classification accuracy and resolve state overlap issues. During training, facial datasets are processed to extract critical features, and a state‐action value table is constructed to guide decision‐making during testing. This reinforcement‐driven learning enables the system to dynamically update its policy based on the most rewarding actions, improving adaptability and performance. Experimental results demonstrate that the proposed approach enhanced traditional models in terms of recognition accuracy, classification efficiency, and training speed. Integrating optimized feature selection and policy learning mechanisms makes the model a promising solution for real‐time and resource‐efficient face recognition applications.