Facial recognition is a key area of artificial intelligence (AI) with applications in security, biometric authentication, robotics, surveillance, healthcare, and commerce. Its widespread adoption is driven by ease of deployment and rapid data processing. However, performance is challenged by factors such as pose variations, occlusions, lighting conditions, and facial expressions, affecting accuracy and reliability. This paper presents a novel approach to enhancing facial recognition accuracy under difficult lighting conditions. The method integrates multiple feature extraction techniques with a convolutional neural network (CNN) model. Advanced feature extraction methods, including Histogram of Oriented Gradients (HOG), Scale-Invariant Feature Transform (SIFT), Oriented FAST and Rotated BRIEF (ORB), Accelerated-KAZE (A-KAZE), and Gabor filters, are utilized alongside traditional techniques like the Canny Edge Detector and Harris Corner Detector. These extracted features are then processed through a CNN, evaluated with different activation functions and optimization algorithms to identify the most effective combination for improved facial recognition performance.