Efficient rendering and display of photorealistic light field images

How to improve the rendering efficiency and realism of light field images remains a major challenge. This paper proposes a rendering approach to efficiently generate photorealistic light field images. Firstly, the next-generation model optimization technique is used to transform the complex high-polygon model (HPM) into the low-polygon model (LPM). Then, the rendering speed is accelerated by using the LPM and geometric perspective coherence. Next, the normals of the HPM are baked onto the LPM using model baking techniques, and the high-precision light field images are rendered using LPM and the normals of HPM. This process can reduce 95% of the model data, thereby increasing the rendering speed by more than 10 times, while maintaining the rendering quality comparable to the HPM. Secondly, nearly half of the triangles are further decreased by using a two-step backface culling technique, thereby doubling the rendering speed again. The rendering speed is greatly improved using the GPU-CPU hybrid architecture. Thirdly, to enhance realism, anisotropic illumination effects are realized by using the perspective coherence of lighting, and vivid shadows are rendered utilizing an image-space shadow mapping technology. The results indicate that the proposed method can increase the rendering speed by more than 20 times and maintain the image's SSIMs exceeding 0.9 and PSNRs above 30 dB. Finally, a light field display system based on integral imaging was built, and high-resolution and attractive 3D images with a wide viewing angle were synthesized using the light field images and display system.