2D/3D Surface Mapping Based On Interest Points Matching Constraints

Realistic rendering techniques have been widely used in computer aided design,virtual reality,film stunt production and 3D online game development.In order to display the realistic 3D models on the computer,2D/3D surface mapping is used to generate the realistic graphic images of the 3D virtual scenes.Surface mapping is established one-to-one correspondences between an image and a model or two models.For matching the best correspondences,we can compute minimization of a measurement.Surface mapping is an important technology of realistic rendering technique.3D face shape fitting and 2D face replacement are challenging problems in computer vision.Within graphic engineering application,it can be attributed to how to find the appropriate mapping problem.In addition,Interest points detection is a primary task of mapping.This paper presents a method of interest points detection based on bilateral filtering.This method uses bilateral filtering to preserve the features while removing the noise in the mesh.Different from the previous work of choosing a circle or a sphere as a local neighborhood,a new method is proposed for choosing a local neighborhood based on tangent plane.The experimental results show that the interest points detected by this method are satisfied with human interests.In order to apply interest points to shape matching and distinguish 3D models,the method for detecting interest points based on multi-scale bilateral filtering is proposed.Firstly,the method constructs a set of meshes with different densities.For each mesh,based on the computed saliency of the mesh vertex,the bilateral filtering is used to remove the noise of the mesh saliencies both considering the spatial closeness and geometric similarities,and the global contrast to normalize the saliencies,and then the interest points are extracted based on the normalized saliency.Finally,this method gathers and clusters all interest points,and the centers of these clusters are treated as the final interest points of the 3D model.The experimental results validate the effectiveness of our method for detecting 3D interest points.We also test the potential for our method to distinguish 3D models.This paper presents a 3D morphable model fitting method based on interest points matching constraints.This method can be used to fit a 3D face model with a single image,and realize alignment between interest points and 2D landmarks automatically.Firstly,initialization parameters are used to calculate the average face model.Then,edge evolution is used to align 2D landmarks and 3D interest points.The landmarks and cheek edge information is calculated to update face shape and pose parameters.Finally,a 3D face model is fitted.The automatic alignment between 2D landmarks and 3D interest points relies on the optimized parameters,which solves the problem that the contour disappears when the parameters change in 2D/3D non-rigid alignment.Finally,this paper presents a method for 3D mapping based on 3D morphable model fitting method.This method projects two different 3D facial models into the 2D plane and maps 2D graphics based on 2D dense correspondence.Firstly,the planar parameterization is used to parameterize a 3D facial model to the 2D plane.Then,according to the known correspondences between the two graphics,we compute the affine transformation matrix between the two 2D graphics and cluster the points which start from the known point pairs following one-to-one mapping rules.Finally,the results of dense correspondences between the points can be used to match two 3D facial models.This method not only realizes the 3D facial models surface mapping,but also can be applied to face replacement and output the realistic images.