An Intelligent Facial Recognition Framework Using Stacked Autoencoders Integrated With Convolutional Neural Network Architectures

Abstract

Facial Expression Recognition (FER) is an important study in the field of computer vision, which focuses on identifying human emotions through facial images automatically and, therefore, improving human-computer interaction. Nevertheless, the current FER methods are usually characterized by the weakness of reliance on facial landmarks, high-dimensional feature representation, and declining resistance to noise and environmental changes. To counter such issues, this research paper suggests a smart facial recognition algorithm that combines the Convolutional Neural Network (CNN) and Stacked Auto Encoder (SAE) to effectively learn features and classify them. First, the pre-processing of the images is carried out through Gaussian filtering to remove the noise to enhance the quality of the image. Then, a discriminative feature set is formed with the help of Grey Level Co-Occurrence Matrix (GLCM) and feature reduction is performed with the help of Singular Value Decomposition (SVD) in order to reduce redundancy. The refined features will be further learnt by the SAE-CNN architecture to obtain improved representation learning. Lastly, an optimized deep neural network is used to classify data that enhances the accuracy of prediction. Experimental analysis proves that the suggested model performs better than the current methods due to its accuracy, precision, and error rate. SAE + CNN integration greatly increases the feature discriminability and generalization abilities and therefore made the system applicable in real-time FER.