Depth estimation from a single 2D image

Abstract

With the growth of the 3D market, the demand for providing 3D content from already existing 2D content has increased. However, the availability of 3D content is very limited, since estimating 3D structure from a monocular image is a challenging task. In this paper, we propose a new nonparametric learning-based method for 2D-to-3D conversion from a monocular image. Our method follows three stages. First, we select K similar images to the input image from an RGBD database. Then, we infer the depth map using the K selected images and their corresponding depths maps. Finally, we refine the estimated depth map. Experiments on dataset were conducted and comparative evaluations with the state of the art are presented.