支持多风格的图像艺术化快速处理算法  

A Fast Image Artistic Processing Method Supporting Multiple Styles

作　　者：史晓颖[1,2] 汤颖[1] 范菁[1] 

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023 [2]杭州电子科技大学计算机学院,杭州310018

出　　处：《计算机辅助设计与图形学学报》2015年第10期1918-1928,共11页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61173097);浙江省自然科学基金(LY14F020021);浙江省重大科技专项重大工业项目(2013C01112);杭州电子科技大学科研启动基金(GK150201205002/004)

摘　　要：为了使目标图像在用户的控制和编辑下快速学习参考图的颜色和笔刷特征,得到多个参考样图的艺术风格,提出一种支持多风格的图像艺术化快速处理算法.首先以目标图像流场为引导对非等轴纹理合成算法进行改进,学习得到样图的纹理和笔刷特征;然后通过设计和实现基于CUDA加速的并行合成算法,加快样图风格的学习速度;此外,为增强学习结果的可控性,设计完成了用户交互式的风格编辑接口,可根据用户指定的不同样图风格块对目标图像实现区域相关的多风格图像艺术化,提供所见即所得的交互设计体验.风格学习实例的结果表明,与已有的基于学习的风格化算法相比,该算法具有更快的合成速度和更强的灵活性.This paper presents a fast image artistic processing method supporting multiple styles, which makes the target image learn the color and brush characteristics of the multiple reference images fast under user's control. Firstly, this method modified the anisometric texture synthesis algorithm to learn the texture and brush characteristics of the reference images by the guide of the target image's direction field; Secondly, a parallel synthesis algorithm based on CUDA was implemented to improve the learning efficiency; Thirdly, in order to improve the controllability of the learning results, a user interactive style editing interface was designed, which achieved regional related image stylization supporting multiple styles according to the dif-ferent sample style patches specified by the user, and provided WYSIWYG design experience. The experi-mental results show that our style learning method has faster synthesis speed and more flexibility compared with the existing stylization method based on learning.

关 键 词：非真实感绘制 图像艺术化 非等轴纹理合成 CUDA 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]