Light field display is an important true Three-dimensional(3D) display technology. It can provide the observers with rich 3D perception such as binocular parallax, motion parallax, monocular focusing, and so on. Light field display technology has a wide range of applications in various fields such as industrial and agricultural production, daily life, military defense, game and entertainment. It is a platform technology, so it has attracted much attention. High-quality light field display requires a large amount of light field image information. How to quickly acquire light field image information is still an important challenge in this field. The light field information acquisition technique mainly includes two categories one is optical acquisition technique and the other is digital rendering method. Optical acquisition technique can capture light field information of real scenes, but the acquisition system is complex and requires precise calibration. The digital rendering method uses computer technique to generate light field images. It can be combined with computer graphics technology to render special artistic effects flexibly. Model-based Rendering (MBR) and Image-based Rendering(IBR)are two typical digital rendering methods. The MBR method can get high-quality light field images, but the rendering efficiency needs to be improved. The IBR method renders faster, but the image quality needs to be enhanced. Both of these methods suffer from the pseudoscopic problem of 3D images. In order to improve the rendering efficiency and quality of light field images, a rendering algorithm of light field image and display system based on conjugate perspective coherence camera are proposed. The algorithm is called Conjugate Multiple Viewpoint Rendering (CMVR). The depth mapping relationship between a normal perspective camera and a conjugate perspective camera is analyzed. The conjugate perspective cameras are constructed and used to render the light field images without pseudoscopic problem in one step. Therefore, the image encoding process is avoided. At the same time, the perspective coherence between the conjugate cameras is used to reduce the redundant calculation and accelerate the rendering process. Compared with traditional algorithms, the proposed algorithm has strong parallel processing capability and is suitable for existing graphics processing hardware to accelerate the process. The rendering pipeline for light field images is built using a hybrid programming technique of CPU and GPU. The test results show that the CMVR algorithm can achieve good image quality comparable to the rendering pipeline of commercial application program interface OpenGL. The rendering efficiency of CMVR algorithm is higher than traditional single viewpoint rendering algorithm when the viewpoint number is larger. The CMVR algorithm is insensitive to the number of viewpoints, and is especially suitable for the rendering of high-density light field images. And the more viewpoints are rendered, the more obvious the efficiency improvement is. In addition, the algorithm can also be effectively compatible with texture mapping, lighting and other technologies in computer graphics to achieve realistic scene rendering. In order to verify the correctness of the algorithm, a light field display system is built. The display system is mainly composed of a light source, a high-resolution LCD display screen, a lens array and a diffuser screen. The 3D images of virtual scenes with good horizontal and vertical parallax are outputted by using the rendered light field images and display system. Vivid 3D imaging effects are obtained. The proposed algorithm can provide an effective tool for fast rendering of light field images. It can also be used in dynamic 3D display, augmented reality, virtual reality, and other fields.
