Deep learning and machine learning based facial expressionrecognition system

In face-to-face communication, words, tone of voice, and facial expressions are crucial for efficient interaction. It is observed that gestures and facial expressions significantly impact interpersonal relationships. Thanks to emotion recognition from facial expressions, various operations in many fields, especially medicine, education, and security, are facilitated. Within the scope of this study, hybrid approach methods were used to increase the efficiency of studies that performed emotion prediction from facial expressions. This developed application consists of two steps: feature extraction and classification. Within the scope of this thesis study, in the first step, SIFT, SURF, and KAZE features were classified using SVM, XGB, RF, and LR classifiers. In the second step, classification was performed using CNN, ResNet50, and VGG16 methods. In the first stage of hybrid methods, by using deep learning and classical machine classifiers in combination with each other, the features obtained with SIFT, SURF, KAZE, ResNet50, VGG16 were classified using SVM, XGB, RF, LR, and deep learning methods. In the final stage of the hybrid techniques and study, the hybrid feature (HessianSIFT) obtained by building a SIFT descriptor on the critical point locations detected with the Hessian detector was classified with SVM, XGB, RF, LR classifiers. According to the experimental results, the recognition success of emotions detected from facial expressions was determined as 100% using ResNet50 and SVM together.