| Facial expression is a means to express human cognition,emotion and state.Accurate and effective facial expression recognition plays an important role in promoting natural and harmonious human-computer interaction.In recent years,three-dimensional(3D)point cloud facial expression recognition has attracted more and more attention because two-dimensional(2D)facial expression recognition methods are largely limited by illumination and posture changes.This paper studies how to recognize 3D facial expressions based on depth features of 3D spatial point sets.The main research work is as follows:(1)An effective representation of 3D facial expression is studied.In traditional facial expression recognition,the method of artificial designing features is often used,which requires the designer’s rich experience and knowledge.This study first uses the depth features of 3D spatial points to represent 3D facial expressions,making it easy to achieve end-to-end recognition of facial expressions.To further improve the accuracy of this facial expression feature,extract Mesh＿LBP(Mesh Local Binary Pattern)is fused with the depth features of a three-dimensional spatial point set as a feature for recognizing facial expressions.(2)A 3D point cloud facial expression recognition method based on improved Pointnet++network is proposed.Aiming at the Pointnet++ network,this method neglects the rich geometric relations between local points and adds an in-neighborhood geometric attention module to the Pointnet++ network.The method in this article was validated using the Bosphorus database.The data were enhanced by rotation and downsampling of the expression data.The enhanced data set is used as a training set and a test set for training and parameter adjustment experiments.The optimal hyperparameter was used to obtain the recognition accuracy of 87.96%.The experimental results show that this method has a good recognition effect on three-dimensional point cloud facial expressions.(3)A method is proposed to extract deep features from local binary features of 3D meshes using convolutional neural networks,and to fuse them with the deep features extracted by an improved Pointnet++network for facial expression recognition.This method combines point set depth features with deep texture features to improve the accuracy of 3D facial expression recognition,and has achieved good results on the CASIA database. |
