Analysis of facial expressions is an active topic of research with many potential applications, since the human face plays a significant role in conveying a person's mental state. Due to the practical values it brings, scientists and researchers from different fields such as psychology, finance, marketing, and engineering have developed significant interest in this area. Hence, there are more of a need than ever for the intelligent tool to be employed in the emotional Human-Computer Interface (HCI) by analyzing facial expressions as a better alternative to the traditional devices such as the keyboard and mouse.
The face is a window of human mind. The examination of mental states explores the human's internal cognitive states. A facial emotion recognition system has a potential to read people's minds and interpret the emotional thoughts to the world. High rates of recognition accuracy of facial emotions by intelligent machines have been achieved in existing efforts based on the benchmarked databases containing posed facial emotions. However, they are not qualified to interpret the human's true feelings even if they are recognized. The difference between posed facial emotions and spontaneous ones has been identified and studied in the literature. One of the most interesting challenges in the field of HCI is to make computers more human-like for more intelligent user interfaces.
In this dissertation, a Regional Hidden Markov Model (RHMM) based facial emotion recognition system is proposed. In this system, the facial features are extracted from three face regions: the eyebrows, eyes and mouth. These regions convey relevant information regarding facial emotions. As a marked departure from prior work, RHMMs for the states of these three distinct face regions instead of the entire face for each facial emotion type are trained. In the recognition step, regional features are extracted from test video sequences. These features are processed according to the corresponding RHMMs to learn the probabilities for the states of the three face regions. The combination of states is utilized to identify the estimated emotion type of a given frame in a video sequence. An experimental framework is established to validate the results of such a system. RHMM as a new classifier emphasizes the states of three facial regions, rather than the entire face. The dissertation proposes the method of forming observation sequences that represent the changes of states of facial regions for training RHMMs and recognition. The proposed method is applicable to the various forms of video clips, including real-time videos. The proposed system shows the human-like capability to infer people's mental states from moderate level of facial spontaneous emotions conveyed in the daily life in contrast to posed facial emotions. Moreover, the extended research work associated with the proposed facial emotion recognition system is forwarded into the domain of finance and biomedical engineering, respectively. CEO's fear facial emotion has been found as the strong and positive predictor to forecast the firm stock price in the market. In addition, the experiment results also have demonstrated the similarity of the spontaneous facial reactions to stimuli and inner affective states translated by brain activity. The results revealed the effectiveness of facial features combined with the features extracted from the signals of brain activity for multiple signals correlation analysis and affective state classification.
|
![[NJIT logo]](http://archives.njit.edu/vhlib/images/njit_logo.gif){kind=logo}
![[Van Houten Library logo]](images/vhl-top-image.jpg)
