Reading Faces, Recommending Choices: A Systematic Review of Facial Emotion Recognition and Recommendation Sy

Abstract

This review emphasizes the necessity of intelligent monitoring in our technologically advanced environment by examining the confluence of recommendation systems and facial emotion recognition (FER) model built on CNNs. To increase the model's performance and accuracy, it is trained using a combination of facial photos and "Action Units" (AU), which capture the movement of facial muscles. Particularly for less common emotions like disgust, the article emphasizes how important it is to train with real-world imagery. It presents a feasible pipeline that combines CNN training with face identification and shows how CNNs perform better in FER than Support Vector Machines (SVMs). Comparing the proposed DL model against state-of-the-art algorithms, tests on the JAFFE and FERC-2013 data-sets show that it achieves greater accuracy, computational complexity, detection rate, and learning rate. The CNN architecture and the procedure for gathering data-sets are both thoroughly described by the authors in their paper. They recommend utilizing AUs to better feature extraction by capturing minute facial movements. In comparison to earlier models, the final model which consists of eight conventional layers, pooling, and dropout layers performs better and is especially good at predicting happiness and surprise.Future possibilities for research include adding more real-world photos to the training data-set, adding tiny expressions to the faces, and putting the model on a distributed platform for real-time applications. The possible application of Histogram-Oriented Gradient for real-time face tracking and identification in HCI scenarios is also mentioned in the paper.