Ensemble learning-based deep neural network model for face recognition
G. Priyanka, Senthil Kumar Jagatheesaperumal · AIP conference proceedings · 2022
Due to the poor regularization capability of single Convolutional Neural Network (CNN), the performance of face recognition system is severely affected. Also, the input facial image data for recognition is influenced by surrounding information like light, expression, brightness, contrast, pose variations and other factors. This issue is overcome by using ensemble-based feature learning ofCNN and local binary patterns (LBP). It also aids in improving the occlusion-related low pedestrian detection rate. To begin, the LBP operator is used to extract texture features from the face and by using different 10 CNN architectures, different part of facial information is learnt. This is mainly used to improve the performance of underlying networking attributes and thereby it leads to enhanced classification results using the Softmaxactivation function. With the introduced new form of face recognition focused on parallel ensemble learning of convolutional neural network and local binary pattern for feature extraction, the obtained results outperforms the existing state-of-art techniques. Finally, utilizing majority voting, the approach of parallel ensemble learning is employed to obtain the final result of face recognition. The ORL and Yale-B face datasets identification rates climb to 98.1% and 99.2 percent, respectively, for our propose system. The proposed approach is demonstrated in the experiments to improve not only the model's immunity to withstand with different illumination, expression, lighting, brightness and posture conditions, but also the accuracy of face recognition with the poor regularization metrics, which denies in reaching the global solution in the solution space.