Facial Emotion Recognition using CNN and VGG-16

Facial Emotion Recognition (FER) is a technique to recognize one's emotional state using facial expression. However, facial expressions are an effective way to recognize human emotions in diverse contexts, but manual FER challenges as it also depends on one's state of mind. Convolutional Neural Network (CNN) models are applied to analyze facial expressions like happy, Sad, surprised, etc. From vocal information, it is challenging to analyze a person's behavior. This work proposes VGG-16 with a transfer learning model to recognize FER. This process assists in the recognition of the position and motion of facial muscles. FER has various steps in which data is pre-processed, features are extracted, and facial emotions are classified. Machine learning algorithm is being used by different researchers to extract the facial emotions but did not get optimal accuracy. So Deep Learning model is proposed i.e. pattern based to achieve more accuracy than the earlier one. The dataset used in this work is collected from the Kaggle repository, which consists of 35,887 samples, in which 28,821 were used for training and 7066 were used for validation. The results shows that proposed model achieves accuracy of 91 percent which is approx. 20 percent higher than tradition machine learning models.