3D clothes modeling from photo cloned human body

Abstract

An important advantage of virtual reality technology is that real 3D objects including humans can be edited in the virtual world. In this paper, we present a technique for 3D clothes modeling based on a photo cloned human body. Photo cloning is an efficient 3D human body modeling method using a generic body model and photographs. A part segmentation technique for 3D color objects is applied for the clothes modeling, which uses multi-dimensional mixture Gaussians fitting. Firstly, we construct a 6D point set representing both the geometric and color information. Next, the mixture Gaussians are fitted to the point set by using the EM algorithm in order to determine the clusters. This approximation gives probabilities for each point. Finally the probabilities determine the segmented part models corresponding to the clothes models. An advantage of this method is that the clustering is un-supervised learning without any prior knowledge as well as integrating geometric and color data in multi-dimensional space.