Research And Application Of Improved Support Vector Machine In The Field Of Recognition Of Facial Expressions

In today's society, computers are gradually replacing and transcending human abilities across all areas of daily life. Making computers have the same and even higher intelligence as human beings are the goal of many of us struggling for life. Human intelligence is composed of logical thinking and emotional thinking, and the most direct and main way of passing emotion between people is facial expression. In this thesis,the core research content is to optimize the SVM's (Support Vector Machine) parameter selection and apply it to the dynamic facial expression recognition. This thesis has done the following work:The thesis designs an intelligent search algorithm combined with Ant Colony Search, Grid Search and Meshing thought, and apply it to the parameter optimization of SVM for the first time. The Ant Colony Algorithm could combine the fastness, global optimization and rationality in the limited time. The Grid Search could overcome the Premature Convergence of the Ant Colony Algorithm. The idea of Meshing is to ensure the search precision based on the reduction in search complexity. The search algorithm finally obtains a satisfactory result in the four-dimensional parameter optimization of SVM (poly).This thesis designs a cross-platform, real-time facial expression recognition solution. This solution combines different technologies, mainly about Boosted Cascade Classifier, angle correction with eyes' detection, image preprocessing, Gabor Filter,PCA, SVM and parameter optimization. The solution is tested with JAFFE database,and achieves 92.857 percent success rate.In the end, the solution is implemented in the mobile, the actual recognition speed reaches 2 times per second. The application of visual expression recognition in smart phone has also been discussed and studied in this thesis, mainly about mixed expression recognition and expression analysis.