Sparse Primitives for Stylized Image Synthesis

这项研究将为计算机生成的风格化图像创建算法，例如，以绘画或雕刻的风格。用户将能够提供自动转换为高质量风格化版本的照片，要求系统从头开始创建新的风景画，或使用交互式工具创建复杂的纹理图像。 该方法依赖于在空间中分布“基元”;基元表示图像的基本特征，例如折痕、脊或轮廓线。给定基元的分布，算法然后填充缺失的信息以创建没有间隙的全色图像。输出图像在基元附近详细，而在更远的地方不太详细。可以使用大量的图元来创建纹理丰富的图像，或者可以部署较少数量的图元来创建更平滑的着色图像。随着这项工作的成熟，新的虚拟环境的实时动画将有可能在很少的人为干预下，适用于计算机游戏和计算机生成的电影等应用程序。

This research will create algorithms for computer-generated stylized images, e.g., in a painted or engraved style. Users will be able to supply photographs which are automatically transformed into high-quality stylized versions, request the system to create a new landscape painting from scratch, or work with interactive tools to create intricate textured images. The method depends on distributing "primitives" through space; primitives represent basic features of the image, such as creases, ridges, or silhouette lines. Given a distribution of primitives, algorithms then fill in the missing information to create a full-color image without gaps. The output image is detailed near the primitives and less detailed further away. Large quantities of primitives can be used to create richly textured images, or primitives can be deployed in smaller numbers to create more smoothly shaded images. As the work matures, real-time animations of novel virtual environments will be possible with little human intervention, suitable for applications such as computer games and computer-generated films.