Research And Application On3D Facial Reconstruction Methods

Research and Application on3D Facial Reconstruction Methods Abstract:In a global sense, different faces are highly similar. Faces of different individuals share the same main features in roughly the same locations, and their sizes do not vary much. However, locally, face shapes can vary considerably across individuals, genders and race. From the field of computer vision, face is the main method to identify different individuals, while its three-dimensional shape can describe the features better due to its properties, and it is more operational. Therefore the3D shape is a research hotspot.Three-dimensional shape and reflectance provide properties of objects that are invariant to the changes caused by the imaging process including viewpoint, illumination, and background occlusion by other objects. Knowledge of these properties can simplify the process of computer analyze images; allow prediction of appearance under novel viewing conditions. So, a major challenge in computer vision is to extract the useful3D information directly from the images, and in particular, when possible, from a mere single image. In this paper, we use the shading information along with the single facial reference model to accurately recover the3-dimensional shape of a novel face from a single image. Because the ambient light conditions of the image cannot be determined, the shadow may exist in the image, which can influence the results. So, the illumination normalization process is used. We introduce a novel method for shape recovery of a face from a single image that uses only a single reference, this method uses the shading information of input image along with the reference shape and albedo to recover the3D shape of a face.The reference model is used to extract information essential for the recovery process that is unknown a priori. With the information, this method can estimate lighting of input image. Finally, the reflectance function becomes only a function of the unknown surface normal. For convergence of the algorithm and the smoothness of the reconstruction results, we also employ appropriate boundary conditions and regularization terms. Experiments demonstrate that this method can efficiently achieve accurate reconstructions of novel input faces overcoming significant differences in shape between the input and reference individuals and some of the most critical problems in recovering the3D models of unknown individuals.