Deep learning algorithms for facial emotion recognition of streaming video image systems

Abstract

ABSTRACT newlineEmotion recognition is considered as the most challenging and interesting task. This thesis mainly focus on recognizing the emotions from the human faces using 3D, 2D images, and 3D videos. In this thesis, the significant contribution that are obtained due to this emotion recognition process is summarized and its potential future scopes is also discussed. Facial- analysis is considered as a trending research topic in computer vision area as it is used in a number of real-time applications like FER (Facial Emotion Recognition), facial recognition, etc. Recently, this type of FER, facial recognition approaches have received a huge attention due to its developed network architectures. Facial recognition is a prominent process which is used in biometric technique to authenticate the identity of person. However, FER is a real- time automated analysis process which plays a crucial role in identifying the facial emotions from video to develop an interaction interface for human robots. The video based FER (VFER) are now used in various real-time applications. newlineAdvancement in technologies advances the social media platform, therefore the availability of images, and videos increased. Videos contains number of frames, within that thousands of emotions are hidden in each human faces. Generally, the human face express millions of emotions within a fraction of seconds which increase the complexity of existing FER approaches. To eliminate such issue an efficient architecture for video based FER from individual and group is required. Existing approaches extract features from detected faces for emotion recognition, as each emotions contains different features. Therefore extraction of features from human faces improves the performance of recognition process. But these extracted features occupy high dimension which maximize the complexity of FER process. This kind of issue is also highlighted in this work which is eliminated by introducing the wrapper based feature selection approach. With that optimal selected features, th