3D Face Reconstruction From A Single Image Based On SFS And 3DMM

At present,the application of three-dimensional face reconstruction technology has become more and more widespread.Among them,the method of three-dimensional face reconstruction through a single picture has gradually become the mainstream method due to the advantages of fast operation speed and little influence by other factors.This paper based on the existing methods,a method of combining improved 3DMM method and SFS method is used to generate a three-dimensional face reconstruction model from rough to fine.First,select some pictures in the AFLW face database as the face dataset in this paper.For the input image,the face position is confirmed by the Haar cascade method;in order to extract more geometric details of the face,68 points are manually labeled as feature points of the face;and then the labeled face is identified by the model Perform segmentation to remove background noise and reduce picture size.Then,in order to establish the connection between the 3DMM method and the SFS method,a Lambertian model was used as the illumination model.The average face model generated by 3DMM is depicted on the Lambertian body,and then the faces in the input image are modeled to obtain a large number of face geometric vectors and texture vectors.In order to reduce the amount of computation,principal component analysis The dimensionality reduction operation generates an undistorted face model.In order to fit the model more closely to the actual contour of the face and to smoothen it,the parameters are adjusted using the method of face registration.The method is verified under the conditions of multiple lighting,multiple expressions and multiple angles,and a rough three-dimensional face model with clear geometric shapes is obtained.Finally,the SFS method is used to estimate the light source direction,direction angle,and inclination angle to obtain information about the light intensity of the face.The minimization method is adopted to accelerate the calculation speed,and the light intensity information of each face is mapped onto the rough matching model to recover the face model.This method is also verified under the conditions of multiple illumination,multiple expressions and multiple angles.The results show that it makes up for the shortcomings of the rough matching model and can obtain a fine three-dimensional face model.In this paper,face reconstruction experiments are performed on a large number of face pictures in the database,which proves the correctness and generality of the method.