Research And Design For Fine-Grained Expression Classification

Image understanding in semantics has becoming a more and more popular research area of computer vision. The goals of researching in this area are the development of algorithm that could be used in an automatic detection system, understanding of the content of an image precisely, realizing the organizing, management and recycle of the image data efficiently. Researchers from computer vision, cognitive science and machine learning have these issues studied in different aspects. In modern computer vision research, understanding of image content has been divided into three levels: Content analyzing based on perception layer (low layer feature), such as color, texture, shape, contour, motion, time-spatial relation, etc. Image understanding based on cognitive layer (middle layer feature), such as concept and semantic information detection for main areas of image and video as well as objects and scenes, etc. Image analyzing on emotion layer (high layer feature), such as emotion classification, expression classification as well as analysis of aesthetics, etc. are the main levels.Classification of human face expression has been an important issue for a long period of time in the field of human face recognition. Most of the modern researches were done on the basic6prototype expressions or the relative action units’ recognition. Automatic expression classification has been very difficult since the diversity and complexity as well as the subjectivity of the human expressions. As it has its irreplaceable value in human-computer interaction and high layer vision semantic understanding, expression classification has been studied with a massive number of researchers from domestic and international facilities. There have been several amazing results from their research. The most popular classification is based on6basic prototype expressions, which are surprise, disgust, fear, joy, sadness and anger. However, these6basic expressions can hardly have the real expressions described. There are certain limitations in both theories as well as applications for expression classification.In summary, according to modern psychology theories, this thesis has the fine-grained expression classification studied and it analyzed the up to date research progress in emotion classification and expression classification as well as the relative psychology concept. We defined30fine-grained expression classes and built a large expression database for the classification. Data features have been analyzed precisely and the concept of fine-grained emotion classification has been defined. Based on all of the above, experiments with different features based on modern research method have been done. Main work in this thesis is:first, modern research progress on expression classification has been thoroughly analyzed, based on which, summary and analysis on ways of expression classification has been posted. Furthermore, a large database was firstly built for fine-grained expression classification and further research. Thirdly, we studied the feature extraction and analysis of expression classification then proposed an effective method to solve fine-grained expression classification issue. From numerous experiments, the effectiveness of proposed algorithm has been proved.