Research On Improving And Optimizing Of3D Reconstruction Based On Single Image

3D reconstruction is a very important research field of computer vision, and there is a broad foreground for its applications in the modern society. The Shape From Shading algorithm is getting the3D shape information from single2D image, which can lay a good foundation for3D reconstruction. The grey information of objects is a key basis in3D reconstruction, therefore, how to using grey information to get the3D characteristics is an important and difficult task. In this paper, we study how to improve and optimize the3D reconstruction algorithm based on single image. The main works of this dissertation are summarized as follows.1) We study the classical SFS algorithms, analyze the illumination model, and discuss the various influence factors in the process of reconstruction; and then we discuss the four classical algorithms:the minimum method, evolutionary method, linear method and local analysis algorithm;2) In order to overcome the shortcomings of the Shape From Shading (SFS) algorithm in items of clearness and smoothness of silhouette of reconstructed objects, a modified SFS algorithm is proposed in this paper. Firstly, we make sure the primary factors affecting reconstruction by analyzing the Lambertian illumination model. Secondly, we perform image preprocessing and then extract object silhouette by edge detection to separate the object and background. After that, we perform image restoration and background smoothing. Finally the object is reconstructed using the SFS algorithm. Theory and simulation results show that, compared with the traditional algorithms, the performance of the modified algorithm can be improved efficiently, which can enhance the accuracy of the reconstructed shape and the continuity of the silhouette of reconstructed objects, as well as reduce the reconstruction errors effectively.3) Due to the high computational complexity and the excessive processing time, an algorithm based on Sobel operator for3D reconstruction is presented in this paper. It applies Sobel operator to compute the grey gradient and then uses numerical integration to obtain the depth information, the cubic spline interpolation is applied to get more depth points so that it can not only reduce the computing complexity, but the performance of reconstruction also be improved. Simulation results show that the presented algorithm can greatly enhance the surface continuity and smoothness.