Research On Camera Arrangement Algorithm In Light Field Rendering

Digital media have influenced and changed human life. Over the last two decades more and more media are produced, processed, stored and transmitted. During the same time, researches and technologies from computer graphics, computer vision, multimedia and related fields enabled the development of new types of media. For example multi-view video, which provides a better sense of reality compared with traditional media. Multi-view video system uses information captured by real cameras to render novel views based on view synthesis technologies. There are two kinds of view synthesis technologies: model based rendering(MBR) algorithm and image based rendering(IBR) algorithm. MBR needs accurate geometry information to build a geometry model of the scene which is hard to realize in practice. IBR uses images of the scene to build model which are easier to acquire and manage. So IBR algorithm has attracted a lot attention. Light field rendering(LFR) algorithm is the core of IBR. During the capturing stage of LFR, a huge number of images are captured by real cameras. The more the amount of the capture cameras, the better the rendering result of the virtual views. What is more, different arrangement of capture cameras also leads to different rendering quality. How many cameras do we need to capture a given scene? If the amount of capture cameras is limited, where can we put them to get the best rendering result? We have done some work to solve these questions about camera arrangement in light field rendering.First, when the camera position is optimal, the rendering error of the virtual views is minimal. By analyzing the classical camera arrangement algorithms, we find it important to find a relationship between camera arrangement and rendering error of the virtual views. We define a mathematical concept named weighted effective area as the medium between camera position and rendering quality. Then we build our own camera optimization model and use improved clustering algorithm to solve the optimization problem. Virtual view rendering result proves the validity of our algorithm.Then, by analyzing the spectral support of the light field we get the minimum sampling rate for rendering light field without aliasing. Minimum sampling rate can help determine the minimal number of the cameras. By the virtual view rendering experiment we found the camera arrangement algorithm can help to reduce the number of the sampling cameras. At last we present a sampling frame based on the camera arrangement algorithm we proposed.