A combined method based on CNN architecture for variation-resistant facial recognition

Identifying individuals from a facial image is a technique that forms part of computer vision and is used in various fields such as security, digital biometrics, smartphones, and banking. However, it can prove difficult due to the complexity of facial structure and the presence of variations that can affect the results. To overcome this difficulty, in this paper, we propose a combined approach that aims to improve the accuracy and robustness of facial recognition in the presence of variations. To this end, two datasets (ORL and UMIST) are used to train our model. We then began with the image pre-processing phase, which consists in applying a histogram equalization operation to adjust the gray levels over the entire image surface to improve quality and enhance the detection of features in each image. Next, the least important features are eliminated from the images using the Principal Component Analysis (PCA) method. Finally, the pre-processed images are subjected to a neural network architecture (CNN) consisting of multiple convolution layers and fully connected layers. Our simulation results show a high performance of our approach, with accuracy rates of up to 99.50% for the ORL dataset and 100% for the UMIST dataset.