A Fast Head Reconstruction Algorithm based on Single Photo

PubDate: November 2021

Teams: Jinneng Holding Coal Industry ；University of Electronic Science and Technology of China,

Writers: Zhen Han; Ye Li; Lei Wu; Li Zhan; Guangqiang Yin

PDF: A Fast Head Reconstruction Algorithm based on Single Photo

Abstract

With the development of Virtual Reality and Augment Reality technology in recent years, the requirement for reconstruction of people based on photos increases gradually. At present, the practical 3D face reconstruction technology is mainly based on the 3D Morphable Model (3DMM) method. However, the collection and production of the 3DMM model are complicated, and the 3DMM model only has a facial part model. As a result, it cannot reconstruct a complete 3D human head and is difficult to apply to actual engineering. To solve those problems, we propose a manual 3D human head reconstruction method, which based on 3DMM. Compared with the traditional 3DMM model production method, the method proposed in this paper reduces the difficulty of production. It adopts the common 3D format so that it can be developed and used by mainstream 3D editing software. In addition, it simplifies the size of the model to make it easy to deploy on mobile devices. Traditional face reconstruction can achieve face reconstruction by directly using the face of the photo as the texture. For head reconstruction, we first made a fixed full-head map, than design algorithm to match the color of the head map with the skin color of the face. Finally, we calculate the affine transformation to transform the face of the photo into the head map and achieve smooth fusion. Aiming at the actual application scenarios, we design the illumination normalization algorithm to deal with the uneven illumination of the face photos, and improves the reconstruction efficiency of the head under the condition of uneven illumination. Experiments show that the above algorithm can reconstruct a realistic 3D human head. The average reconstruction speed on the i7-4790 CPU is only 700ms.