• This work investigates recognizing face images using a novel original deep learning model called ScaleSyncNet , and we believe it aligns well with the scope and focus of your journal. • A proposed face recognition system is presented in this work using ScaleSyncNet model. A newly developed algorithm has been employed for feature to acquire on the most significant and powerful features that describe the face details. Moreover, our model achieves high recognition accuracy in minimum time and complexity. The suggested system provides excellent recognition accuracy with little computing effort and complexity in uncontrolled contexts, with an average accuracy of 100% in 254 ms, as demonstrated by the comparison of experimental findings with current approaches, and time, respectively, which show an excellent performance. In this study, a hybrid deep model (ScaleSyncNet) is proposed in order to recognize the facial images in unconstrained environment under varying conditions like occlusion, pose variations, and low quality. A multi-scale convolutional feature extraction is included in the proposed model architecture with synchronized temporal modeling through Gated Recurrent Units (GRUs) in order to obtain all types of facial features including local and global. Three variant datasets are employed to evaluate the proposed system: MUCT, CASIA-WebFace, and ORL, and the acquired accuracy from the deep model ScaleSyncNet surpass traditional models like VGG16, AlexNet, and ResNet, showing a 100% accuracy on both MUCT and ORL datasets and 99.9% accuracy on CASIA-WebFace, while maintaining high computational efficiency of a training time ranging from 254 to 653 milliseconds, thereby demonstrating its robustness, versatility, and ideal candidate for real-time face recognition scenarios.
Kadhim et al. (Sun,) studied this question.