| 3D digital presentation of commodities provides an intuitive shopping experience for customers in E-Commerce.This raises the challenging issue of reconstructing real-life products with high quality and low cost,and texture mapping plays a crucial role in the 3D reconstruction of real products.The background of this article is to use low-cost,consumer-grade scanning hardware to reconstruct the product in 3D,with a lot of noise in the process.As the state-of-the-art texture mapping methods cannot stably generate high-quality texture maps for noisy input data,this paper presents a common seamless texture approach for Multi-view 3D reconstruction.We present a seamless texturing method that consists of generation,correction and fusion of the seams.Firstly,a novel energy term based on Markov random field is proposed to generate reasonable seams located at low frequency area of images.Then,we employ a global seam correction process to further eliminate the remaining visible seams by optimizing the camera pose and picture distortion iteratively.Finally,global illumination adjustment and local Poisson fusion are performed to generate continuous illumination of texture.Experimental results show that the proposed method can provide seamless and high-quality texture maps even for noise data. |
