Image-based Garment 3D Display Technology

Clothing has always occupied a large proportion of the e-commerce,especially in online retailing.At this stage,display methods of clothing goods are mainly based on two-dimensional images.Although this traditional approach can meet the basic requirements of demonstration,compared with the three-dimensional approaches,it always lacks realism and cannot deliver the interactive experience between users and clothing goods.This paper mainly focuses on imagebased 3D clothing reconstruction.Firstly,we discussed the key technical steps and related algorithms in image-based 3D geometric reconstruction,and subsequently verified the related algorithms in reconstruction such as stereo vision,Poisson reconstruction and texture mapping.Besides,a point-cloud fusion strategy was proposed to avoid the dense point cloud missing and mesh reconstruction error caused by clothing self-occlusion.To this end,the point cloud reconstructed in batches was registered and fused in terms of the Iterative Closest Point algorithm.In addition,the influence of different image-shooting scheme and different octree depth in Poisson reconstruction has been evaluated.Finally,a pseudo-3D display method based on the combination of photo stream and face-swap technology was proposed for personalized display and efficiency improvement.The experimental results shows that,1).the proposed multi-view 3D clothing reconstruction pipeline could effectively recover the target point cloud from the sequence images.The overall topological structure of the generated point cloud clearly restores the spatial shape of the original clothing;2).Using the nearest point iterative algorithm to fuse the point cloud can effectively solve the reconstruction error in mesh reconstruction and improve the realism of the 3D clothing;3).A better 3D clothing model can be obtained when the number of input images is greater than 60,along with the "hemispherical" shooting scheme and setting the octree depth in Poisson reconstruction to 11.In addition,the feasibility of personalized display scheme based on single image presented in this paper has also been verified.