A Modeling Method for the Human Body Model with Facial Morphology. (December 2021)

• Record Type: Journal Article
• Authors: Fang, Naiyu; Qiu, Lemiao; Zhang, Shuyou; Wang, Zili; Wang, Yang; Gu, Ye; Tan, Jianrong
• Is Part Of: Computer aided design. Volume 141(2021)
• Publication Date: 2021-12
• Subjects: Human body model -- Facial morphology -- 3D landmark localization -- Weak-rigid local registration -- Virtual try-on; Computer-aided design -- Periodicals; Engineering design -- Data processing -- Periodicals; Computer graphics -- Periodicals; Conception technique -- Informatique -- Périodiques; Infographie -- Périodiques; Computer graphics; Engineering design -- Data processing; Periodicals; Electronic journals; 620.00420285
• DOI: 10.1016/j.cad.2021.103106
• Languages: English
• ISSNs: 0010-4485

Abstract

Abstract: The 3D human body model for virtual try-on requires not only accurate human body shape estimation but also detailed facial morphology representation. The facial shape and expression of the traditional 3D human body model are neutral with a sense of visual alienation. In addition, the traditional 3D face model lacks complete human body support with a visual sense of suspension. Thus, a new modeling method for the human body model with facial morphology is proposed in this paper. Our model has an accurate representation of human body shape and brings people a sense of self-identity with the model. In this paper, we proposed a 3D landmark supplement method for the facial region of the human body model, which is taken as the registration data. We proposed the weak-rigid local registration between the human body and face model by restraining the sparse 3D landmark position and global shape, which includes the translation rotation and scaling of the face model. We proposed the head cut and packed ring generation method based on the Bezier curve that further solves the interference between components. In the 3D landmark localization experiment, the supplemental landmark average error of the face models on the HFC dataset is approximately 0.2 mm. In the registration experiment, in the case of high precision, the convergence of the restraining landmark position is verified, and the average number of steps is 1125. In the case of real-time application, some measures are also provided to complete the registration in less than 10 s; and compared with the NR-ICP registration, our method has better performance locally and on noncorresponding structure registration. In the model evaluation experiment, the universality and application in virtual try-on are demonstrated, and our method has better performance in virtual try-on compared to SMPL-X or MTC. Highlights: Supplement the 3D landmarks on the facial region of the human body model. Propose the weak-rigid local registration with non-corresponding structure method. Cut the head region of the human body model to remove the interference further. Generate the packed ring to fill in the blank zone based on the Bezier curve.