Enhancing Emotion Recognition in Facial Images with Deep Learning

Emotions in humans are a naturally occurring state of mind. Because of its great diversity and the hazy relationship between emotions and facial expressions, facial recognition is a difficult field of study. Face recognition has become a key part of biometric identification technology with the rapid development of this field. It simplifies numerous corporate applications. Recognition of facial features is becoming more and more popular. However, there are still certain shortcomings in the present methods. In this project, we proposed a face emotion identification technique for determining the emotion of people. We discuss the issue of distinguishing between the face and the emotion. Face emotion recognition is extremely difficult since some facial features are obscured. One of the main objectives of researchers working on emotion identification is that lots of people are not happy, or depressed to reduce the percentage of this emotion giving them a solution to be happy. The process begins by collecting regions from an already trained deep convolutional neural network, primarily focusing on areas like the eyes and forehead. Next, a Bag-of-Features model is used to analyze the feature maps from the final convolutional layer, providing a simplified representation compared to the fully connected layer found in a traditional CNN. Finally, a Multi-Layer Perceptron (MLP) is utilized during the categorization step. The experimental results demonstrate a high level of success in facial recognition with ResNet50 showing exemplary performance with an accuracy of 98.43% and precision of 71.46%.