Unveiling Emotions: A CNN Approach to Facial Emotion Recognition

Abstract

Introduction: Facial emotion recognition is a unique field of research that can be exploited in numerous areas like in the prevalent areas getting AI, gaming, marketing and medical care. The mission of this project is to design the facial emotion recognition system with convolutional neural network (CNN) technique. The CNN algorithm has many advantages especially in respect to the image-processing tasks including facial expression identification because it is able to extract high-level features from images using these models. One of the projects through which we apply this project involves processing input images of human facial emotions and training the pretrain models with datasets. The algorithm provides the computer with the capability to immediately identify animals direct from their face movements. Data augmentation techniques can be implemented using the libraries like Keras to improve the performance of CNN model thus eliminating the need for additional data. These methods are consisting on recovery the training data by applying transformation like rotation, horizontal flip, and offset. They increase the model’s general capability to recognize emotions and enhance the precision. The main purpose of the project is to contribution to the facial emotion recognition recognition domain , and to achive quality and fast recognition by mean of employing the CNN algorithm . To capitalize upon deep learning and image processing capabilities, the system will be enabled to identify and categorize emotional expressions by human faces with a high level of precision.

Objectives: Create a CNN model that can accurately identify various emotions through facial expressions such as happiness, sadness, anger, surprise, fear, and disgust.

Methods It is important to gather a wide range of facial images with emotions labeled to create a diverse dataset. We used FER2013 dataset. Before training the Convolutional Neural Network (CNN), images may need to be resized, pixel values normalized, and the dataset augmented for better model performance. Training the CNN involves feeding batches of images into the network, calculating the loss (the difference between predicted and actual emotions), and adjusting the model's parameters using optimization techniques like Stochastic Gradient Descent (SGD). Key metrics for assessing facial emotion recognition models include accuracy, precision, recall, F1-score, and analyzing confusion matrices.

Results Our findings confirm that facial expressions can indeed distinguish between different emotions, as we anticipated. The utilization of image processing and artificial intelligence, specifically the convolutional neural network, played a crucial role in achieving these results. 

Conclusions: The CNN model performs better in recognizing happy, neutral, and surprised expressions compared to identifying angry, sad, and fearful emotions.