[3DSAp2/3Dp2] 3D and Hyper-realistic Displays and Applications 2

[3DSAp2/3Dp2-10] Object-centered View Synthesis using Learning-based Image Inpainting

*HONG-CHANG SHIN¹, Gwangsoon Lee¹, Ho min Eum¹, Jeong-Il Seo¹ (1. ETRI (Korea))

キーワード：HMD, mobile, motion pallax, view synthesis, image inpainting

This paper presents object-centered view synthesis technique using multilayer concept. we divide the image into multiple layers based on depth information and then provide different motion parallaxes for each layer depending on the depth. When the disocclusion region appears due to motion parallax, the uncovered region is filled by using learning-based image inpainting.