基于OneCut和共享抠图算法的自适应衣物目标抠取

摘要

The clothing database is an important part of virtual fitting system. In order to better meet the need of generating clothing database in image-based virtual fitting system, the adaptive clothing target matting method is proposed in the pa-per, which is based on OneCut and shared matting. First, the paper utilizes segmentation algorithm named as OneCut to segment the cloth image from background. Second, the trimap is automatically generated by mathematical morphology meth-ods. Third, shared matting algorithm is used to obtain the clothing foreground and alpha image used to implement virtual try-on. It is clearly proved by the proposed experiment that it can make use of less interaction to produce the more ideal gar-ment foreground image in real time.%衣料库是虚拟试衣系统中非常重要的一部分。为了更好地满足基于图像的虚拟试衣系统中衣料库生成的需要，本文提出一种基于OneCut和共享抠图算法的自适应衣物目标抠取方法。首先，利用OneCut算法将服装图像从背景中分割出来。然后，使用数学形态学方法自动生成三分图。最后，通过共享抠图算法得到服装图像前景和用于虚拟试穿的alpha值。实验结果表明，能够利用较少的人工交互，实时地生成较理想的服装前景图像。