2.0 Literature Review
Face detection is a computer technology that will identify human faces in arbitrary images and human faces basically have the same basic configure appearance such as two eyes above a nose and mouth. After the computers have successfully on detecting the faces, there are more researches have done in face processing include emotion recognition.
2.1 Face Acquisition
In this process, user’s faces are acquired in order to extract out the facial features from cluttered background. In Robust Real-time Object Detection (P.Viola ,2002), the authors used AdaBoost algorithm to detect the frontal view of faces rapidly. The system able to detect the face from background quickly and ...view middle of the document...
Although the speed of detecting the face by just depending on skin colour is very fast, in reality most of the time this technique is very inaccurate because of different people has different skin colour and also the brightness of environment. In article illumination Independent Colour Based Face Detection (Kovac, Peer & Solina, 2003), the authors proposed their algorithm illumination independent to reduce the problem and improve their skin colour detection technique. However, colour based detection is not that recommend if the system are going to develop further such as feature extraction and emotion classifying in this project. Since the system is unable to extract the feature points of face if the system only relies on skin colour. So, more technique will be discussed in the following section.
2.2 Face Feature Extraction
During this step, the feature extraction can be categorized into two categories which are geometric and appearance. Geometric approach has been used in Recognizing Action Units for Facial Expression Analysis (Y.Tian, 2011), the authors used multi-stated models to extract the geometric features such as used three states to describe lip: open, closed and slightly closed and two stages to describe eyes: open or closed. These parameters will be served as an input of action units (AU) to recognize the facial expression.
While in Cascaded Classification of Gender and Facial Expression using Active Appearance Models (Y.Saatci, 2006), the system used appearance method for facial emotion recognition and overall the achievement was between 64% to 94%. It seems that the appearance approach has the difficulty to differentiate which objects should be in the same class. Take in example, the face albedo should be separated from the class. Appearance models also have the difficulty such as segmentation and normalization is hard to process when the image is present in mutual shadows and occlusion. In order to solve the problem, some researcher combined both approach to improve the system performance.
In article Capturing Subtle Facial Motions in 3D Face Tracking (Wen,Z & Huang, T.S, 2013), the author combined both of the approaches. The authors proposed a ratio image based to remove appearance features which are dependency of a people’s face albedo and since this method is independent of face’s albedos, it can adapt to new user. Appearance features in face region are extracted instead of points. The system will calculate the average for each region. As a result, there are eleven regions has found which has shown in Figure 2.1.c. In this system, the author also applied EM-based algorithm to adapt to lighting conditions.
Figure 2.1 combination of geometric and appearance approaches; a) original image b) geometric approach c) facial regions for extraction
Although we can enhance the edges of the image by filtering before extract the face features but sometimes the face appearance can be affected by various factors such as lighting condition,...