The accurate matching between clothing and human body is one of the key problems in virtual fitting system.Unlike previous 3D or 2D image matching,we propose a new virtual fitting algorithm based on 2D image transformation which has a better body fit.It uses Ki-nect somatosensory equipment to capture the color image,the binary image of human body,the coordinate of human joints and specifies the image of clothes and the corresponding coordinate of the joints on the clothes.The binary image of human body and the image of clothes are divided into multiple blocks twice according to the coordinate of human joints and the corresponding coordinate of the joints on the clothes. Then it fills the image of clothes into the image of human body block by block with affine transformation.The algorithm realizes the simula-tion of the stretching texture of clothing with a strong body fit.Results show that the virtual fitting method by image transformation owns more authenticity.%衣服与人体的精确匹配是虚拟试衣系统中的关键问题之一.与以往的3D模型匹配、2D图像匹配等方法不同,提出了一种新型的基于2D图像变换的更具有人体贴合性的虚拟试衣算法.算法使用Kinect体感设备捕捉到的彩色图像、人体二值图像、人体关节点坐标,和指定的衣服的图像与衣服上对应关节的坐标.通过人体关节点坐标和衣服上对应关节的坐标,以过关节点最短探测线,分别将人体二值图像和衣服的图像按人体骨架控制区域进行一级分割,接下来对所有一级分割结果使用射线探测的二分方法进行二级分割.并根据相对应的模块,把衣服的图像通过仿射变换逐块填充至人体图像上.通过实验实现了衣服拉伸质感的模拟,有较强的人体贴合性,结果表明通过图像变换的方式进行虚拟试衣更具真实性.
展开▼
