Abstract Surface splatting has proven to be a good approach to render point-based models. The technique keeps the simplicity of point based models with a high quality rendering due to local filtering between the samples. However, silhouettes and sharp features are noticed as high frequency variations on image-space. Thus, filtering cannot hide artifacts to appear next to these areas without an explicit approach, mainly on bad sampling conditions as low density of samples. In this paper, we present a new curved splat approach, called here quadric splat, which improves rendering of a model near silhouette and sharp features. These quadric splats may be placed on a surface using any sampling process applying our proposed error metric. We also propose a simple method to detect splats near silhouettes on object-space. This approach allows an efficient silhouette detection on GPUs. Our silhouette detection was applied on a surface splatting pipeline aiming to render quadric splats only where their effect is more noticed, but the technique is independent and can be applied even for non-realistic rendering of splat-based models.

中文翻译：

---

基于 Splat 模型的改进的轮廓渲染和检测

摘要 Surface splatting 已被证明是渲染基于点的模型的好方法。由于样本之间的局部过滤，该技术保持了基于点的模型的简单性和高质量渲染。然而，轮廓和锐利的特征被认为是图像空间的高频变化。因此，过滤不能在没有明确方法的情况下隐藏出现在这些区域旁边的伪影，主要是在样本密度低等不良采样条件下。在本文中，我们提出了一种新的弯曲 splat 方法，这里称为二次方 splat，它改进了接近轮廓和锐利特征的模型的渲染。可以使用应用我们建议的误差度量的任何采样过程将这些二次方碎片放置在表面上。我们还提出了一种简单的方法来检测对象空间上轮廓附近的碎片。这种方法允许在 GPU 上进行有效的轮廓检测。我们的轮廓检测应用于表面 splatting 管道，旨在仅在效果更明显的地方渲染二次方 splats，但该技术是独立的，甚至可以应用于基于 splat 模型的非真实渲染。