Research And Implementation Of High Robust Video Face Expression Recognition System

Expression is one of the important ways humans express emotion.The facial expression recognition system allows the computer to have the function of autonomously understanding human emotions.It is of great significance for achieving harmonious human-computer interaction.Facial expression recognition system has high practical value and broad market application prospects.It is one of the hot research topics in artificial intelligence today.Although facial expression recognition technology has achieved great development now,there are still some problems in practical use: because of the different focuses of the research,the existing video facial expression recognition methods have fast recognition speed but poor robustness;the existing single-image facial expression recognition methods have good recognition effects,but the recognition speed is not good.Therefore,the paper has studied and designed an expression recognition system that is based on video and has strong robustness so as to satisfy the actual conditions of use.The specific work is as follows:(1)The research status of facial expression recognition is analyzed.In view of the problems in practical use of expression recognition systems,a highly robust video facial expression recognition system is constructed.(2)Adaptive acquisition and preprocessing of video facial expression images are implemented which improves the robustness of the system in terms of noise,occlusion,lighting and scale changes.(3)The traditional 60-site ASM facial key point location method is improved: according to the characteristics of facial expressions,a re-planning of the key points of the human face is re-planned and using the ASM+structure method instead of the traditional pure ASM method to locate the key points of the face.Experiments show that although the number of key points obtained by the new calibration strategy decreases,the expression information they contain is more effective.The ASM+structure method is much less computational than the pure ASM positioning method.They lay the foundation for improving the robustness and real-time performance of facial expression recognition systems.(4)The improved facial key point location method and Gabor transform are integrated which implements Gabor feature extraction and dimension reduction for facial expression images.The differences in key points at different sites were analyzed.For the key points of the upper and lower half faces,different scales and multi-directional Gabor transforms are performed respectively which achieves further organic dimensionality reduction of expression features.Experiments show that this method not only utilizes the advantages of Gabor transform in terms of robustness,but also solves the problem of Gabor feature data redundancy through dimensionality reduction.The facial expression feature extracted by this method reasonably retains the effective facial expression information of the key points of the face.The system's real-time performance and robustness are improved.(5)A 1-to-multiple and 2-to-2 combination SVM classifier is designed and implemented for 6-category facial expression recognition.Experiments show that this classifier has a 3.02% higher recognition rate than the one-to-many SVM classifier when it recognizes non-occlusion face images.And when the classifier is combined with the expression feature extraction method proposed in the paper,the whole system is robust to partial occlusion and has the characteristics of fast operation speed.