This study revisits the global mean pooling layer (GAP) proposed, and sheds light on how it allows Deep Residual Neural Network to have a remarkable localization capacity, work with the same technique, but in three different forms, with the change of the type of network ResNet50, because it also works by the system (GAP) which is used to obtain the final convolutional layer representing the characteristics of the human face. In our experiment, we used “yalefaces” dataset with the addition of many blurry images to have the efficiency rate of this technique to detect the face. The experimental results from the proposed methods and the face recognition rate for the “yalefaces” dataset are validated, and we compared with recent techniques. The face recognition rate of the used dataset based on this network is 99%, which shows the effectiveness.
Hassene Seddik hasn't uploaded this paper.
Let Hassene know you want this paper to be uploaded.
