Improved Performance of CNN Classifier-Based Face Recognition Using HOG Features Extractor Against Variant Conditions

Abstract

Face recognition is a critical task in various applications, such as surveillance, security, and humancomputer interaction. Recently, Convolutional Neural Networks (CNNs) have emerged as a powerful tool for face recognition, outperforming traditional methods in many scenarios. However, the performance of CNN-based face recognition systems can be affected by various conditions, such as pose, illumination, images quality and occlusion. This paper aims to investigate the performance of CNN-based face recognition classifier under different challenging conditions such as low resolution, face poses, low light condition, ages and various facial expressions. This research focuses on the evaluation of the effectiveness of CNN-based face recognition systems and evaluates their performance as a classifier on a combination of the Celebrities face dataset and the Indian Movie Faces Dataset (IMFS) have been trained in order to obtain varying levels of illumination conditions, images quality, face poses and facial expressions. For face detection, the Haar Cascade face detection technique is utilized for this purpose. The facial features extraction is a vital process in this system. Therefore, Histogram of Oriented Gradients (HOG) is employed as facial features extractor method. The findings suggest that by carefully designing the CNN architecture and incorporating techniques to improve information propagation, the performance of the face recognition system can be enhanced, even under adverse conditions. The obtained results show an improved level of accuracy in which the proposed system achieved 96%.