The Research Of 3D Face Reconstruction Algorithm Based On A Single Photo

With the rapid development of economy and the progress of science,human have a kind of higher pursuit in the material and spiritual.In the game animation,film and television plays,medical beauty,video communications,information security and many other aspects,human beings are no longer satisfied with the visual sensory experience of two-dimensional world,which lead to the birth of 3D movies,television and game animation.The 3D face modeling has become a focus of attention in the field of scientific research and practical engineering because of its intuitive display effect and wide application.Compared with the traditional 3D face reconstruction based on three views,multi views and video streams,it is the focus of the present study that the minimum amount of data needed to be reconstructed the face mode based on a single image.However,the facial structure is extremely complex and huge differences between different human faces.It is difficult to obtain the depth data of face directly through a single face photo,which is more challenging to achieve.Therefore,it is of great research value and practical value to reconstruct 3D face model based on a single photo.The two following tasks are carried out based on a single image reconstruction:This paper proposed an algorithm of feature point depth recovery based on feature fusion.Firstly,the 3D face database is preprocessed in order to obtain the corresponding two-dimensional face image.Secondly,the coordinate information of facial feature points is extracted by explicit shape regression.Then,according to the obtained feature points,the facial feature regions are divided by Delaunay triangulation.Compared with the similarity between the recovery face and the database,we can recover the depth information based on geometric features.What's more,the local texture region is divided by feature points and the LBP histogram of each region is statistically analyzed.In order to recover the depth information based on local texture feature,the distance of LBP histogram is measured between the recovery face and the database.Finally,by way of improving the precision of feature points,the least squares method is needed for feature fusion.In this paper,we proposed an algorithm of feature point depth recovery based on feature fusion that feature is easy to extract and the algorithm complexity is low.The experimental results show that the depth information of facial feature points is more accurate.This paper proposed two kinds of texture mapping algorithms.The texture mapping method based on Delaunay triangulation of constrained refinement need to compute the mapping relationship between the 2D face image and the 3D model by the feature points.In order to realize the further Delaunay triangulation of constrained refinement,we perform scatter interpolation in feature regions.The mapping is used to establish the face model and achieve texture mapping.The texture mapping method based on radial basis interpolation train the radial basis network by the known feature points and build the specific radial basis network.Then,the non-feature points are interpolated and mapped by the network to reconstruct the real face model.Compared with the traditional method,the proposed method does not need a large number of human-computer interaction,and uses a small amount of data to reconstruct a more realistic face model,which is suitable for application in engineering.