Novel View Synthesis Based on View-dependent Texture Mapping with Geometry-aware Color Continuity(<Special issue>VR Psychology VI)

書誌事項

タイトル別名
  • Novel View Synthesis Based on View-dependent Texture Mapping with Geometry-aware Color Continuity

この論文をさがす

説明

Novel view synthesis (NVS) is renowned for its variety of applications, and recent techniques are based on an image-based approach, which uses multiple input images to synthesize novel views, with the help of 3D geometry. View-dependent texture mapping is one of the most well-known techniques of this kind. However, it sometimes suffers from blurry or duplicated edges when the geometry is inaccurate because it blends several input images for alleviating inconsistent lighting conditions among input images. This paper presents a view-dependent texture mapping-based method for NVS that leverages geometry-aware color continuity in texture selection. Our color continuity constraint suppresses noticeable texture boundaries in resulting novel views without blending, while our method relaxes the constraint on steep geometric changes. Since such geometric changes are often followed by surface property changes, our new color continuity constraint can lead to more natural novel views. We formulate our texture selection by an energy minimization problem and solve it with the graph cut algorithm. Our experimental results have demonstrated the superiority of our method.

収録刊行物

詳細情報 詳細情報について

問題の指摘

ページトップへ