Point-based View Synthesis from RGB-D Images

Abstract

This paper proposes a method generating novel views from a set of RGB-D images captured by static multiple depth cameras. The novel views were generally synthesized by back-projecting the real 3D points onto the virtual image plane located at the required viewing position. In this case, empty pixels and holes were visualized when the 3D scene is zoomed-in by the user. To solve this problem, the proposed method constructs 3D structures from point clouds converted from depth images, and renders novel views from the constructed point clouds based on RGB-D images. In the experiments, the proposed method can reduce excessively blurred images, compared with the previous point-based rendering methods.