In Augmented Reality, it is important that a virtual object is superimposed based on environmental information to improve the user experience. When the virtual object is made of metallic material, it is also important to generate the reflection of the surrounding scene on the object in order to reproduce its material appearance. In this study, we propose a method to reproduce the metallic material appearance of a virtual object from limited information on the background in an input image using an image transformation network. The network receives an input image consisting of a background scene and the normal map of a virtual object and transforms the normal map into the image of a virtual object with a metallic material appearance by generating environment reflection. A dataset consisting of input images and target images was created using 3DCG rendering software. Experimental results showed that the proposed network could generate natural environment reflection and reproduce a metallic material appearance in an unknown environment.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Reproducing Metallic Material Appearance by Generating Environment Reflection Using Image Transformation Network

  • Koki Masuda,
  • Taishi Iriyama,
  • Takashi Komuro

摘要

In Augmented Reality, it is important that a virtual object is superimposed based on environmental information to improve the user experience. When the virtual object is made of metallic material, it is also important to generate the reflection of the surrounding scene on the object in order to reproduce its material appearance. In this study, we propose a method to reproduce the metallic material appearance of a virtual object from limited information on the background in an input image using an image transformation network. The network receives an input image consisting of a background scene and the normal map of a virtual object and transforms the normal map into the image of a virtual object with a metallic material appearance by generating environment reflection. A dataset consisting of input images and target images was created using 3DCG rendering software. Experimental results showed that the proposed network could generate natural environment reflection and reproduce a metallic material appearance in an unknown environment.