High Dynamic Range Environment Sequence Sampling

Rendering realism is a key point for more multimedia applications. Customers also pursue high-quality display. Real scene illumination is used for the rendering of virtual scenes to improve quality. Because high dynamic range environment images with high solution store a lot of information, the lighting computation for one vertex on the objects in the 3D scene is very large. Environment sampling is a classic technology for raising efficiency and saving time. The techology extracts proper samples from HDR environment images to approximate original ambient light, which brings high efficiency, space saving and promising real-time rendering. Compared to static sampling,sequence sampling approaches acquire coherence, less flickering artifacts and higher accuracy. Previous work treats an environment sequence as a whole. Although it can get good temporal consistency, more time and storage space are required. And some other methods are sampling environment sequence with static sampling methods, that may lead flickering artifacts. Here, we employ temporal superpixels to generate segmentations. Temporal superpixels have unbalanced importance for rendering though irregular shape of light areas and temporal coherence can be maintained. To solve this issue, we apply an adaptive merge-and-split scheme to adjust light regions to receive a more balanced importance distribution. Compared with state-of-the-art methods, our approach can attains better rendering image quality for frames. Meanwhile, experiment results show decent coherence of sequence rendering.