Filling holes in manifold digitized 3D meshes using image restoration algorithms Emiliano P´ erez 1 , Santiago Salamanca 2 , Carlos Cerrada 3 , Pilar Merch´ an 4 and Antonio Ad´ an 5 Abstract— A method for filling holes in manifold 3D meshes based on a 2D image restoration algorithm is expounded in this work. To do that, data must be converted to a suitable input format: a 3D to 2D transformation is executed by projecting the 3D surface onto a grid. Therefore, the proposed algorithm starts by a first stage of holes identification. Then, a meaningful mesh portion is chosen for each hole. Afterward, the suitable plane of projection must be computed to get the range image of the mesh portion. Later, the image restoration algorithm is applied to the range image. Finally, an inverse transformation 2D to 3D is performed and the new produced data are merged with the initial mesh. The result is a robust algorithm which works correctly with several kind of holes and for different sizes of them. I. INTRODUCTION The three-dimensional modeling is useful for a wide range of industrial applications and is carried out usually by increasing the use of range sensors. Complete reconstruction of the 3D surface from the in- formation gained from a range sensor can be divided into four stages: the acquisition of partial views, the registration or alignment of these views, the integration of all views in a single representation, Finally, the processing of the mesh. This last stage is very important to get a model which presents a geometrical and topological correctness. One of the treatments applied in the processing stage is the holes’ filling. This algorithm aims to obtain a watertight mesh, completely closed, which is useful in many applica- tions: manufacturing systems, rapid prototyping, characteri- zation of surfaces, object recognition, etc.. In the literature can be found a great variety of techniques of filling holes in meshes. In [15] they propose a review of several of them that can list classified into two main categories: Methods in which the filler is an implicit process in creating the 3D model. Methods in which the filler is a separate process to create the 3D model. 1 E. P´ erez is with Escuela T´ ecnica Superior de Inform´ atica. Universi- dad Nacional de Educaci´ on a Distancia. Madrid, Spain emiliano at issi.uned.es 2 S. Salamanca is with Escuela de Ingenier´ ıas Industriales. Universidad de Extremadura. Badajoz, Spain ssalaman at unex.es 3 C. Cerrada is with Escuela T´ ecnica Superior de Inform´ atica. Universi- dad Nacional de Educaci´ on a Distancia. Madrid, Spain ccerrada at issi.uned.es 4 P. Merch´ an is with Escuela de Ingenier´ ıas Industriales. Universidad de Extremadura. Badajoz, Spain pmerchan at unex.es 5 A. Ad´ an is with Escuela de Inform´ atica. Universidad de Castilla la Mancha. Ciudad Real antonio.adan at uclm.es From all the proposed techniques, we can stand out among the first type methods, usually based on point clouds, those methods which interpolate the original data using alpha shapes ([2], [10]), crusts ([1], [8]) or spheres ([3]), and methods that fit a set of radial basis function (RBF, radial basis functions) to the data as [9], [5]. As regards the second type, which are more numerous, the following stands out: [7] is a representative filling method of those using implicit functions. Here it is implicitly calculated a function of distance, defined in the vicinity of the hole to fill. Then it applies a diffusion process that extends the surface along the volume and closes the hole. Nooruddin and Turk in [14] propose a method that starts with the voxelization of the mesh for filling holes and repairing ill-defined areas (double walls, faces intersection ...). In [19] they propose a technique based on texture synthesis methods that take advantage of context information to undertake the filling of the holes. The advantage of this method is that it provides very realistic results in certain types of surfaces. Wang and Oliveira in [20] take a point cloud as input and generates an intermediate representation consisting of a mesh of triangles upon which an edge detection is applied to determine the holes. After that, they take a group of neighboring points on these edges and an interpolation is performed using the MLS (Moving Least Squares) technique [12]. In [21] vertices of the environment of the hole’s boundary are used such as interpolation centers to define a local implicit surface, which serve, using Radial Basic Functions (RBF) to interpolate the contents of the hole. In [13] they present a method that uses all polynomial fitting techniques, and has the advantage of preservation of features of the surrounding surface of the hole. The method proposed in [4] performs a projection process of the mesh and, thus, carries out the stuffing in 2D. Despite the diversity of techniques exposed, it can not be said that there is a valid method for filling any hole that may appear in a mesh. In this paper we propose a method of filling holes in meshes, applied to complete objects. This approach would fit within the second category mentioned above and in particular in those using traditional techniques used in 2D photographs. The main idea is based on the use of 2D image restoration techniques (image inpainting), for filling holes in 3D. Therefore, the algorithm will require a process of adaptation between the two type of data: 2D and 3D. In section II, the method will be described in general Proceedings of the 2012 IEEE Intelligent Vehicles Symposium Workshops ISBN: 978-84-695-3472-4 1