3D Facial Expression Features Selection And Recognition Based On Bi-objective Elitist Strategy And Semantic Knowledge

At present,the key problem for 3D facial expression recognition is:The facial expression recognition is based on the low-level visual features of images,however,the human understanding of image is based on high-level semantic knowledge,there are completely different.This problem has not only affected the face expression recognition accuracy and robustness seriously,but also blocked the popularization and application of facial expression recognition.On the other hand,Features extraction and selection as the first important process of facial expression recognition,the reasonable filter directly affects the accuracy of the subsequent facial expression recognition.Therefore,the research to solve the "semantic gap" and "feature selection" problems of 3D face expression recognition is a very meaningful subject.In the view of the above problems,a method of 3D facial expression recognition,which is based on bi-objective elitist strategy and semantic knowledge,has been processed.Major works and innovations of this paper are listed as follows:?1?The features selection which is based on bi-objective elitist strategy has been proposed:The most of feature selection algorithms now are based on reduce the vectors' dimension.However,feature selection is actually a process of dual targets control,reduce the number of features,and also ensure the similar degree between same expressions and the difference between different expressions are both large.Therefore,in order to solve these problems,this paper has proposed The Improved Non-Dominated Sorting Genetic Algorithm ?.Successfully convert the single object selection problem to bi-objective,then proposed to improve the elitist strategy of the original algorithm to solve the local convergence and prematurity,and promoted the features selection effect.?2?The 3D face expression recognition based on semantic knowledge has been proposed:Every face expression models in the BU₃DFE database have six basic expressions,this paper treat every six basic expression as a six dimensions vector of semantic knowledge.This paper established the mapping relationship between feature vectors and semantic vectors by KCCA algorithm.In this way,the algorithm can connect the low-level visual features and high-level semantic knowledge together and narrow the "semantic gap".?3?The experiments have been done trough the 3D facial expression database.The whole recognition rate of 86.89%has been obtained.The theoretical analysis and experimental results both show that this method has a better recognition rate than the others.