Depth Recovery From Defocus Images

Recovering depth information of 3D object from images is an important issue of computer vision. Currently, the most concern and research methods are recovering from double or multiple view, depth from focus images, depth from defocus images. Compared to optical measurement using active method, camera as a passive method has been widely used due to its protability. But the images which camera obtains are 2D, without the depth information. How the depth information from one or more 2D images effectively recovery becoming the difficult task in 3D computer vision of areas.Depth from defocus images recovers depth of object in scene using defocus as the cue. Since this method has been proposed, it attracts many home and abroad researchers to carry out long-term and many work which have a great of progress. With the partial differential equation in the image processing applications, the depth from defocus once again become the focus of researchers, which is based on partial differential equation having the solid foundation of mathematical theory and a clear physical background.In this thesis, the main work combines partial differential equation, functional and variational method. Starting from physical theory of optical defocus of imaging and deduces the relationship between the depth and the diffusion coefficient of point diffusion function (Gaussian convolution). Further, clarify the Gaussian convolution is the heat equation (known as the diffusion equation) of the basic solution, therefore, the heat diffusion of partial differential equation can describe the physical characteristics of defocus imaging process. Heat diffusion equation has a widely research and application in science and engineering, including the isotropic diffusion and anisotropic diffusion. Thesis employs the calculation of anisotropic heat diffusion model defocused imaging process; gives the "energy" functional corresponding to the discussion problem; analysis the relationship between the diffusion coefficient and the edge which depth is jumping, excluding the influence of edge which depth is not variant on the energy functional; employs L1 norm measure of diffusion coefficient gradient as a constraint functional items; recover the depth information of object by solving the minimum of constraint energy functional.In order to verify the validity and accuracy of algorithm, the experiments are on simulated and real images in this thesis. The experiment results show that the presented algorithm which recovers the depth from defocus images is valid. Then, on the simulated images the thesis uses absolute error and relative error calculation method. Compared with the least squares algorithm, the presented using L1 norm measure has small error. End, the thesis has a discussion between the camera parameters and algorithm error. The result of discussion shows that the presented has a widely application.