The 3D Facial Expression Recognition Based On Face Surface Normal Density Images

The research of facial expression recognition has always been the focus of many researchers.The topic is not only of practical value but also a huge challenge.The research plays the important role in achieving the harmonious development of human and machine.Besides,the technology can be used in robot manufacturing,safe driving,clinical medicine,education platform,computer games and so on.Throughout the past few decades,researchers have focused on the research of 2D's facial expression recognition.Therefore,the technology based on the 2D facial image recognition has become increasing mature,but at the same time,this technology has encountered the bottlenecks.The recognition rate is difficult to further enhance,and the recognition result is inevitably affected by light,pose,occlusion,background,perspective etc.The essential reason is that 2D facial expression image is only a projection of the 3D face,which losts a lot of information.In order to overcome these problems,this paper presents a new method of 3D facial expression recognition.This method is robust,simple and efficient,compared with the conventional 3D facial expression recognition method.3D facial expression recognition includes image acquisition and preprocessing,expression feature extraction and expression classification.The main contents of this paper are as follows:Firstly,we need to acquire and preprocess the face expression image.We take Minolta VIVID 700 camera to acquire the images.The camera can achieve 3D face images in 0.5 seconds or so.The Minolta VIVID 700 camera as the image acquisition equipment,the camera broke the ordinary camera,can obtain 3D face image to achieve 0.5 seconds or so of time.On the one hand,the 3D face data obtained by Minolta VIVID 700 contain the data of the human face depth,which improves the recognition rate,compared to the 2D data obtained by the ordinary camera.On the other hand,this paper puts forward a new concept which is defined as the image plane.According to the concept,we can transfer 3D facial data into 2D face data.The virtual face plane is obtained by the method of minimum variance and minimum distance.Then,we need to extract the expression image feature.In this paper,according tothe fact that the facial expression is changed with the facial muscles and bones,we analyze the facial muscles and bones by the face place,and get the face normal density image.It can quickly distinguish facial expressions.Finally,the effective fixed points are extracted from the normal density image as the feature points.Finally,we need to classify the expression images.This paper takes the method of Mahalanobis distance and support vector machine based on Mahalanobis distance to classify the expression images.The combination of Mahalanobis distance and support vector machine based on Mahalanobis distance can improve the recognition rate,but also reduce the interference between samples.This paper fully describes the realization of 3D facial expression recognition technology.The research focuses on the 3D data feature extraction methods and classification methods.The experimental results show that the method of deriving the normal density image from the face surface can be used to distinguish all kinds of facial expressions.The classification method has the characteristics of high accuracy and short time consuming.This paper lies the foundation for the research work and determines the direction of future research.At the end of the paper,it presents the limitation of this paper and the new target.