Content-aware processing of vector images

在计算机图形学中，二维图像通常以两种方式之一表示。栅格图像是彩色样本的网格；它们是日常计算机用户最熟悉的类型，因为数字照片是以这种方式表示的。然而，在专业图形设计和插图的世界中，矢量图像也是无处不在的，也是重要的。矢量图像更像是一幅画：它由一系列形状组成，这些形状被填充并用纯色或逐渐变化的颜色勾勒出来。因为它们将图像中的对象表示为精确的几何图形，所以矢量图像具有可以任意缩放而不会造成任何质量损失的优势。矢量图像在电脑桌面和移动设备的屏幕上很常见(首先对它们进行光栅化，这样它们就可以显示为像素)，在平面广告和标牌上，以及在服装上。在许多视频游戏中，它们也被用来表示角色和背景。 今天，已经有了复杂的算法，可以对数字照片进行智能分析和转换。你的相机可以检测到你的拍摄对象何时微笑，并通过多次独立曝光以各种方式合成复杂的照片。Adobe Photoshop等桌面图像编辑软件可以通过对背景对象进行潜移默化的更改来动态重塑数字图像，无缝更改照片的内容，甚至更改照片中个别对象的颜色或材料。 相比之下，用于编辑矢量图像的工具仍然运行在相当低的抽象级别。这些工具更直白：它们根本不能理解正在编辑的图像的结构和内容。这是不幸的，特别是考虑到矢量图像包含大量未被利用的几何信息，例如对象的大小和形状以及它们之间的关系。 这项研究探索了利用矢量图像的内在结构来表示和处理矢量图像的方法，让平面设计师在矢量图像上拥有一些摄影师现在享受的栅格图像的权力。我将探索对矢量线艺术的分析，以支持更智能的着色操作，这些操作可以用于漫画创作。我将研究形状和纹理的表示形式，它们允许从小的示例自动创建大的、不重复的纹理块。我还将开发动态重塑矢量图像画布的技术，而不会扭曲该图像中对象的形状。 这项工作承诺重振当代对矢量图形能力的研究，并为专家设计师和插图画家提供新的工具，使他们更有效、更具表现力和更具创造力。

In computer graphics, two-dimensional images are typically represented in one of two ways. Raster images are grids of colour samples; they are the kind that everyday computer users are most familiar with, because digital photographs are represented in this way. In the world of professional graphic design and illustration, however, vector images are also ubiquitous and important. A vector image is more like a drawing: it is made up of a sequence of shapes that are filled in and outlined in solid or gradually changing colours. Because they represent objects in an image as precise geometry, vector images have the advantage that they can be scaled arbitrarily without any loss in quality. Vector images are common on computer desktops and the screens of mobile devices (where they are first "rasterized", so that they can be displayed as pixels), in print advertising and signage, and on clothing. They are also used to represent characters and backgrounds in many video games. Today there are sophisticated algorithms that can perform intelligent analysis and transformations of digital photographs. Your camera can detect when your subject is smiling, and compose complex photographs in various ways from multiple independent exposures. Desktop image editing software like Adobe Photoshop can dynamically reshape digital images by making imperceptible changes to background objects, seamlessly alter the contents of photographs, and even alter colours or materials of individual objects in photographs. In comparison, tools for editing vector images still operate at a fairly low level of abstraction. These tools are more literal: they are simply not as able to make sense of the structure and content of the image being edited. This is unfortunate, especially given that a vector image contains a significant amount of geometric information that is not being exploited, such as the sizes and shapes of objects and the relationships between them. This research investigates ways of representing and manipulating vector images that exploit their inherent structure, giving graphic designers some of the power over vector images that photographers now enjoy with raster images. I will explore the analysis of vector line art in order to support more intelligent colouring operations, which could be used in comic creation. I will look at representations of shapes and textures that permit large, non-repeating blocks of texture to be created automatically from small examples. And I will develop techniques for dynamically reshaping a vector image's canvas without distorting the shapes of the objects in that image. This work promises to revitalize contemporary research into the capabilities of vector graphics, and empower expert designers and illustrators with new tools that make them more effective, more expressive, and more creative.