Facial expression recognition on partial and whole face images with machine learning methods

Facial expressions are important for interpersonal communication also play an important role in human machine interaction. Facial expressions are used in many areas such as criminal detection, driver attention monitoring, patient monitoring. Therefore, automatic facial expression recognition systems are a popular machine learning problem. In this thesis study, facial expression recognition studies are performed. In general, the applications of facial expression recognition can be grouped under two topic in this thesis: analysis of partial facial images with classical machine learning methods and analysis of whole facial images with deep learning methods. In the first application, classification of the facial expressions from facial images was performed using only eye and eyebrows regions. This approach is different from the studies which are studied facial expression recognition in the literature and high success rate was achieved. With this approach, proposed system is more robust for under facial occlusions and mouth motion during speech. Further, according to our experiments, the generalization ability of the proposed system is high. In this thesis, the rest of the facial expression recognition applications was developed with whole face images using deep learning techniques. One of the proposed methods is segmentation of facial parts with CNN. After segmentation process, facial segmented images were obtained. With this segmented images, personal privacy is protected because the segmented images don't include any personal information. Also, the success rate of the classification was increased with combining original raw image and segmented image. Because; eyes, eyebrows and mouth are crucial for facial expression recognition and segmented images have these areas. Therefore, the proposed CNN architecture for classification forces the earlier layers of the CNN system to learn to detect and localize the facial regions, thus providing decoupled and guided training.