Facial Expression Reconstruction Based On Affective Computing

The research on face modeling and animation technology has a history of more than 30 years.Nowadays,the development speed of human-computer interaction,film and television,virtual reality and other fields is accelerating,and people are paying more and more attention to research fields such as face modeling and animation.Since a three-dimensional face contains more identity information than a two-dimensional face,and the applicable scenes are more abundant,the field of computer vision pays more attention to the exploration of threedimensional face reconstruction technology.The acquisition of real face data in 3D face reconstruction requires a high cost,and the efficiency of traditional algorithms in 3D face reconstruction is not ideal.In addition,the changes in the geometric structure of the human face and the appearance of facial expressions are complex and diverse,and only a small amount of information can be obtained from a single two-dimensional photo,and it is difficult to reconstruct a three-dimensional human face with rich expression changes.Based on this,this paper conducts research on 3D facial expression reconstruction.The specific content is as follows:(1)Realize the facial micro-expression recognition algorithm based on convolutional neural network.Due to the great progress in the depth and breadth of research in the field of machine vision,the exploration of facial expressions has also evolved from the most popular expressions to complex and changeable expressions,from static images to dynamic images.As an important part of emotional computing,micro-expression recognition can effectively extract and recognize facial expression information.The micro-expression recognition algorithm designed in this paper first preprocesses the data set,then uses CNN to extract the time-scale features of the micro-expression video sequence,and finally uses the support vector machine(SVM)to effectively recognize the micro-expression.From the experimental data,it can be concluded that the false recognition rate of this method is less than 0.1%,and the recognition time is no more than 5ms.Compared with the traditional method,this method has a good application effect.(2)The algorithm of 3D facial expression reconstruction is realized,and the facial shape,posture and expression features can be reconstructed at the same time.Recovering a threedimensional face from a single two-dimensional image usually uses a CNN-based threedimensional model.Although this model can be reconstructed on the basis of two-dimensional images by regression of three-dimensional deformation coefficients,it lacks facial expression data,so it will be significantly Affect the realism of facial expression reconstruction.In this paper,we perform expression coefficient regression based on the micro-expression information extracted from the face,and propose a face multi-attribute learning method.The microexpression recognized in the emotion calculation is used as input to learn together with the face shape attributes and posture attributes.Resnet101 network is used to reconstruct a facial model with expression.In order to avoid the influence of facial occlusion on the final expression reconstruction as much as possible,the multi-feature fusion scheme is used in the paper to complete the preprocessing of related images,and then control the influence of occlusion on the experiment.Finally,the algorithm is compared with the reconstruction algorithm based on deformation model and image-to-volume pixel.The experimental results show that the 3D facial expression model reconstructed in this paper has great advantages in terms of local effect,realism,robustness and other performance,and the algorithm is low in complexity and time-consuming.