Data-driven 3D Human Head Reconstruction

Data-driven 3D human head reconstruction refers to the reconstruction of the human head(face,hair)using a database(face database,hair database).Face and hair model reconstruction has a wide range of applications in game character modeling,virtual reality applications,virtual fitting,and customized customization.Using professional modeling software(such as Maya,3ds Max,etc.)for avatar modeling requires the user to have in-depth understanding of the software itself and master certain related art knowledge,and it is difficult to personalize customization;Using professional 3D laser scanning equipment costs too high in collecting and rebuilding and it is difficult to spread;Using database to reconstruct is more suitable for promotion and application.Therefore,this paper studies the method of automatic reconstruction of face and hair from a single RGB image,so that an ordinary user can obtain a personalized 3D head model at a small cost.This paper proposes a framework for reconstructing 3D human head models from a single image.Firstly,a preprocessing system for images is designed.The image is automatically segmented using a full convolutional network.The Gabor filter is used to extract the direction of the hair,and a simple manual interaction is used to modify the results.Secondly,the parameters of the FLAME face parametric model are solved by using feature points.The high-frequency information of the face image is used to enhance the detail,and the texture map is complemented by the albedo parameterized model.Finally,a strip mesh hair database is constructed,and the hair in the image is reconstructed by using this database and the information extracted from the image.The experimental results show that the framework proposed in this paper can reconstruct 3D human head model in the case of inputting a single picture,and the result meets the original picture,has a certain sense of reality,contains face and hair,and is suitable for real-time rendering.