Research On B-spline Face Reconstruction Based On Data From Various Sources

3d face model is widely used in application of computer vision,such as security monitoring,virtual reality,film animation and so on.And most modeling methods are based on polygon.In fact,limited by the topology and accuracy of polygon,polygon models can not meet the needs of higher accuracy and ease of use when using data from various sources that are affected by variety of imaging devices,environmental conditions and states of target face.In contrast,B-spline is closer to describe an organic shape like faces.It can provides an accurate interpretation of face geometry.This paper aims research methods of B-spline reconstruction to handle data from various sources:Building general B-spline face model based on a small number of arbitrarily captured images data of one identity;Reconstructing high-precise B-spline face model based on unconstrained face images;Building B-spline morphable model for detailed fitting accurately.The study of B-spline face model reconstruction covers three levels,including general model,detailed model and morphable model.First,we introduce a general B-spline face modeling method,to construct the C~2face surface that overcomes the shortcomings of discontinuities of traditional mesh model in geometric expression.B-spline surface is optimized based on structure opti-mization,which can be constructed on a B-spline surface with physical contours.First-ly,low rank are employed to estimate the aligned 2D structure,to calculate the depth information with SSIM.And then 3D face surface is reconstructed based on B-Spline by transforming control points in space.According to theoretical analysis and experi-mental results,the feasibility and effectiveness of building a unified model of B-spline face is proved.Then,considering the shortcomings of non-detail unified B-spline face model,a high consistency modeling method based on B-spline parameter domain is proposed to reconstruct more detailed face than mesh model based method.On one hand,the mod-eling is completely consistent with the law of geometric optics,and B-spline reduces the distortion during surface deformation.On another,0th-and 1st-order consistency of stereo are formulated based on low-rank texture structures and local normals,re-spectively,to approach the pinpoint geometric modeling for face reconstruction.By combining the two consistency,a practical solution as well as an iterative algorithm is proposed to optimize high-detailed B-spline face effectively.Theoretical analysis and experimental results prove the effectiveness of precise consistency modeling based on B-spline parameter domain.Finally,this paper proposes a parameter T-spline morphable model(T-splineMM)for detailed face fitting in shape subspace,which has a great advantage in fitting un-known source data in contrast with classical 3D face morphable model.In the model,we describe a face by C~2T-spline surface,and propose a morphable model based on topological linear subspace of T-spline face function in parameter domain constructed by Riemann mapping.A fitting algorithm is proposed to optimize coefficients of T-spline control point components along pre-learnt individual and expression subspaces,as well as to optimize the details in refinement progress.Any pre-learnt subspace covers a limited span of face morphing while T-splineMM makes a larger span of morphing subspace.Theoretical analysis and experimental results prove the superiority of the T-splineMM in detailed 3D fitting task.Summary,we research the technologies of B-spline face reconstruction systemat-ically from three levels,including general facial model,detailed model and morphable model from low to high,to solve the key subproblem in face reconstruction task includ-ing consistency modelling and detailed fitting.Theoretical analysis and experimental results prove the feasibility,the efficiency and the superiority of B-spline modeling in precise face reconstruction task.