Descriptive and generative models for 2D vector pattern design

点击翻译按钮获取中文摘要

2D vector patterns (2DVP), also called vector textures, whether they be static or dynamic, are ubiquitous in many computer graphics and multimedia applications such as artistic or scientific illustration, cartography, data visualization, web design, video games, and visual effects in animation. All these information technology domains are of high importance and of strong industrial impact in Canada. Compared to raster textures, 2DVP hold several desirable properties for computer assisted design: a resolution independence that prevents any aliasing artifacts, a compact structure that reduces memory storage, and an object-based representation that allows novel and powerful editing procedures. Despite recent progress, dedicated computer tools to assist the user in the design of such images are limited. One reason is the lack of a straightforward model for the objects' spatial organisation, an aspect that clearly distinguish them from their raster counterparts and other types of vector images. Another part of the explanation comes from stylization effects such as color and geometry effects which are often correlated to the spatial organisation. The overall research objective of our program is to continue fundamental and applied research for the development of novel intuitive representations and design algorithms of 2DVP. We will first conduct a thorough user study of the manual design process of 2DVP. We will collect online and offline data to understand both the temporal and spatial structures involved in such images. Using machine learning techniques to conduct our analysis, this study will give us insights to propose faithful models depending on the application. Based on the user study, we will propose first a new descriptive model of 2DVP. Current approaches are mostly inspired from raster models, not taking full advantage of the object-based nature of such patterns. We will investigate statistical descriptive models of the spatial organisation combined to perceptually-based style features. This descriptive model will be evaluated in an image retrieval context: web photo stocks of vector patterns contain millions of 2DVP for computer graphic designers. Tags and keywords are expensive, subjective, and not adapted to abstract images such as vector textures. We will also propose a complete generative approach that allows the correlated stylization of objects within 2DVP. A strong user-centered and interactive approach will be adopted here. Finally, this program will investigate techniques to assist designers in the production of animated 2DVP. We plan to propose a programmable approach to design affine transformation effects.