Research On 3D Face Expression Recognition Method Based On Multi-attitude

With the development of science and technology,the extensive development of e-commerce,graphics and image processing technology gradually integrate into life and with the growth of human needs,the claim for graphic image processing technology increasingly demanding.Face recognition technology occupies a very important position.At present there are many areas of concern and extensive research.There are very important applications,such as: criminal investigation,rail traffic,online payment,online shopping,access attendance,etc.Expression is rich in emotional information of the human heart.At present,the two-dimensional face recognition technology has matured,but the two-dimensional face data due to the interference of light,attitude and other external factors.Therefore,it has limitations in the study of face expression recognition.However,the three-dimensional face data with an explicit spatial shape characterization and not affected by illumination,view and make-up and other factors.Therefore,in recent years,facial expression recognition research has gradually shifted from two-dimensional data-based research to three-dimensional model-based research.The main work of this paper includes:1)Automatic landmark detectionMost feature extraction on 3D face recognition is based on the existing data in the established key points.Most of the feature points in the database are manually annotated,the accuracy is lacking.At the same time,in order to alleviate the impact of the head posture change on the nose point location,this article first locate the tip of the nose in an efficient way and then perform automatic landmark detection based on the position of the nose tip.In this paper,according to the location of the already positioned nose tip,combined with the local neural field model and the facial texture model to create the facial landmark location we need for facial expression recognition.The proposed method is effective for extreme head posture and normal head posture.2)Create a descriptorIn order to effectively describe facial expression features,based on the predecessors,this paper proposes a multimodal fusion feature descriptor.The GeoTopo operator and the hot core signature and LBP operator are innovatively merged.The fused features can effectively alleviate the effects of the non-stretch deformation of the 3D face model.It is also effective to describe facial landmarks extracted from extreme head poses.Helps to improve the accuracy of the algorithm in the representation recognition process based on the facial motion action unit.3)Classification and identificationIn the recognition stage,for different head positions,for the non skin color race,the muscle movements generated by the typical six facial expressions are consistent.So the facial expression based on the action unit(AU)based reference facial motion coding system(FACS)facial motion coding system is used to identify the facial expressions.For the extreme head position,a three-dimensional face point cloud cylindrical projection method,which is easy to calculate and rotates,is used to transform the problem into a missing data classification problem.Face 3D data is projected onto a head invariant two-dimensional representation.This article not only realizes facial expression recognition based on action unit,but also realizes the comparison of recognition accuracy between different descriptors.