项目名称: 自然纹理的对称正则化优化方法及其应用研究
项目编号: No.61303101
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 吴惠思
作者单位: 深圳大学
项目金额: 25万元
中文摘要: 网络环境为人们提供了触手可及的自然纹理素材,如何从网络纹理图像中自动提取对称正则化程度较高的理想纹理样图,从源头上提高后期纹理合成应用的效果,是研究人员面临的新挑战。本项目研究自然纹理的对称正则化优化方法及其应用,将系统地建立自然纹理的对称正则化优化理论框架。针对自然纹理随机变化的光照、形变和尺度分布,分别给出纹理光照场、形变场和尺度场的对称正则化优化方法,突破传统只能从光照、形变和尺度变化较小的纹理图像中提取样图的局限;基于纹理光照、形变和尺度的耦合性,提出整合光照、形变和尺度的全局对称正则化优化方法,提高样图在光照、形变和尺度定制上的灵活性和多样性,并实现对称感知的全局优化纹理合成应用(包括纹理替换、纹理修复、纹理抽象与压缩等)。本项目将实现从纹理样图自动提取到纹理合成应用的全局对称感知,从而将传统纹理技术从像素级、方块级和特征级的优化控制拓展到对称感知的语义级优化控制。
中文关键词: 纹理分析;对称正则化;最优化方法;样图提取;纹理合成
英文摘要: Internet provides people with ubiquitous natural texture sources. How to automatically extract high quality symmetrized texture exemplars from internet texture images becomes a crucial issue for improving the texture synthesis effects, and has become a challenge in computer graphics. This project will focus on symmetry regularized optimization methods for natural textures and its applications. We will propose a novel symmetry regularized optimization framework for natural texture images. To handle the randomly distributed lighting, deformation and scale fields within natural texture images, we will present symmetry regularized optimization methods for lighting, deformation and scale fields, respectively. We will break the limitation of traditional methods in extracting exemplars from texture images with non-uniform lighting, deformation and scale distributions. Based on the interconnected lighting, deformation and scale of natural textures, we will present a lighting, deformation and scale fused global optimization method to improve the flexibility and diversity of the exemplar extraction. In addition, we will demonstrate several symmetry-aware texture synthesis applications, including texture replacement, texture inpainting, texture abstraction and compression. This project will realize automated extraction of
英文关键词: Texture analysis;Symmetry regularization;Optimization;Exemplar extraction;Texture synthesis