Research Of Cognitive Machine Learning Methods For Facial Expression Recognition

With the rapid development of Artificial Intelligence technology,machines have the ability to "understand" human emotions through face images.The method of performing this ability is facial expression recognition method based on machine learning.The typical applications of facial expression recognition include the human-computer interaction,security,medical health,and so on.There are seven basic categories of facial expression recognition,including anger,happiness,sadness,surprise,disgust,fear,and neutrality.Therefore,facial expression recognition can be realized by machine learning methods such as classifiers.In real applications,the facial expression images to be recognized may contain complex background,human appearance and age in the images are different,and the boundary between different expression categories may be blurred.These potential problems lead to the difficulty of improving the accuracy of facial expression recognition and the lack of robustness in practical applications.There is still a big gap between the performance of existing facial expression recognition methods and human recognition ability.The main reason is that the cognitive ability of human is not simulated enough.Machine learning is to learn from human beings.Every time the cognitive rules of human beings are modeled,machine learning methods have achieved remarkable originality.Therefore,the main work of this thesis is to apply cognitive rules to machine learning,to propose some new machine learning methods,and then to use them for facial expression recognition.The main work are as follows:(1)It points out the difficulties that the current machine learning methods faced,analyzes the inertial thinking principle of machine learning,and then puts forward a cognitive-based machine learning framework,in which the main problems to be solved in the framework are discussed.The framework provides the novel ideas for proposing some new machine methods.(2)A new complexity perception classification algorithm is proposed and applied to facial expression recognition.This algorithm improves the classification accuracy of facial expression recognition significantly by discriminating the feature distribution of samples.Experimental results verify the effectiveness and universality of the complexity perception classification algorithm.(3)A new sample awareness personal method(SAP)for facial expression recognition is proposed,which uses Bayesian learning method to select the best classifier from the whole world,and then uses the selected classifier to identify the emotional category of each test sample.SAP chooses the best classifier from the given basic classifiers to classify the testsamples.SAP is more in line with human cognitive law and has personalized classification ability.Experimental results show that SAP is more effective than any basic classifier in facial expression recognition.(4)Machine learning is easy to form illusory inertial thinking,but at present all machine learning methods do not considere it.In this paper,a machine learning method based on reverse thinking is proposed.Reverse thinking is used to overcome the illusion inertia thinking,which improves the generalization ability of machine learning method.Experimentals show the effectiveness of the proposed method.The proposed method is universal so that it can be applicable to any machine learning method,especially for those methods on difficult data sets such as unbalanced data sets,where machine learning can easily form the illusory inertial thinking.