Facial Expression Recognition Using CNN with Keras

Facial Expressions are an essential feature of non- verbal communication, as we move towards the digitization the human computer interactions plays a vital role. The emotional changes results change in the expressions. This paper elaborates development of Deep Convolutional Neural Network Model using tf.keras for building and training Deep Learning Model. The aim is to classify facial image into one of the seven face detection classifiers using open CV and one of its classifiers for drawing the boundary box around the face to detect the correct expression. For training the CNN models we have used 48x48 grey-scale images from Kaggle's ICMP 2013-Fecial Expression Recognition (FER) dataset The FER dataset is divided into two folders called test and train, further divided into separate folder each containing one of the seven types of FER dataset. To understand the spread of the distribution of the class data augmentation method is used to generate minority classes. To reduce over fitting of the models, dropout and batch normalization is used. We are using atom optimizer and softmax activation function as it is a multiclass classification problem. It is a categorical cross entropy and matrix that we are training this for accuracy based on the parameters to evaluate the performance of the developed CNN model by looking at the training epoch history.