Many applications such as virtual reality computer games physics simulations and medical applications use 3D scenes Reconstructing 3D scenes from 2D images is common and can be done in one of two ways: (1) from multiple images and (2) from a single image While many well-developed methods exist for 3D reconstruction from multiple images a lack of essential information means that 3D reconstruction from single images remains a challenge We propose a method for reconstructing 3D scenes from a single image under two conditions To complete the shapes of the objects in the reconstructed scene we need the first condition: 3D template models similar to objects in the input image are available To synthesize a complete texture we need the second condition: the input image contains duplicate objects; these objects are repeated in slightly different directions For example take an image of a lotus which has many similar petals that lie in slightly different directions We can reconstruct the 3D scene from this image if we find a 3D model of a petal from the Internet and synthesize the texture of each petal by considering information from other petals In the present study we use template models to reconstruct 3D scenes because 3D model files are now more publicly available on the Internet than in the past The entire work is divided into three sub-problems: (1) model fitting (2) mesh deformation and (3) texture synthesis We can reconstruct the 3D scene after resolving these problems
3D Scene Reconstruction from a Single Image with Duplicate Objects Using Template Models
傑程, 張. (Author). 2017 8月 24
學生論文: Master's Thesis