Emotion analysis from facial expressions using image processing techniques

Facial expressions, which play a major role in interpersonal communication, provide information about emotions, thoughts and mental states. This interaction among people is aimed to be used in human-machine relationships with the advancements of technology. Therefore, understanding and learning human emotions by machines has become a significant issue. However, it is necessary to process facial expressions rapidly and as accurately as possible. In this study, 7 emotional states of "anger, disgust, fear, happiness, sadness, astonishment and natural" have been analyzed. There are basic steps to be taken before classifying these 7 emotional states, including; determining the face, cleaning and normalizing the image, extracting the features from the image, tracking and classifying the changes in the face. A study was conducted on a dataset of 35887 facial images to determine facial expressions. Using this data set, 7 different facial expressions were classified. In this study, 48x48 images in gray format were applied to our proposed CNN model and their performance was evaluated. In the study, a convolutional neural network (CNN) was proposed to automatically classify facial expressions in the FER2013 dataset, and the results were obtained by testing the "man" and "sgd" optimization functions separately as optimization functions. While the accuracy rate was 69% in the proposed model using the "Adam" optimization function, the accuracy rate was 92% in the proposed model using the "sgd" optimization function. Thus, it has been determined that the accuracy rate of the sgd optimization function in the proposed model is higher than the accuracy rate of the man optimization function. Keywords: Image Processing, OpenCV, Facial Expression Classification, Deep Learning, Artificial Neural Networks, CNN