首页|基于薄板样条插值的弯曲笔触神经绘画与风格化方法

基于薄板样条插值的弯曲笔触神经绘画与风格化方法

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
近年来，图像生成技术取得了令人瞩目的发展，目前的图像生成方法大多以像素填充的方式生成图像，缺乏艺术家逐笔绘制的过程，使其在局部笔触细节与纹理上与真实艺术作品有所不同。神经绘画致力于模仿人类艺术家以画笔为单位，逐笔绘制的作画过程。现有的神经绘画方法大多使用贝塞尔曲线或者笔触模板进行仿射变换来模拟真实笔触。然而，贝塞尔曲线纹理的缺乏以及仿射变换的线性性质，导致生成的笔触在纹理或者形状上存在较大的限制。为了更好地模拟真实笔触的纹理与形状，本文提出了新的基于薄板样条插值的弯曲笔触参数模型，通过对真实笔触模板先后进行弯曲与仿射变换，可以生成更加真实、多样的笔触图像。此外，本文提出了层次化的笔触优化方法，将整幅图像分解为由大到小的多个笔触，能够有效提升模型对图像整体架构与局部细节的绘画能力。最后，本文将提出的方法拓展至风格迁移中，实现了较好的风格迁移效果。定性与定量的实验表明，本文所提出的新的笔触模型与优化方法在神经绘画及风格化任务中都超越了已有的最佳模型。
Curved-stroke-based neural painting and stylization through thin plate spline interpolation
In recent years,there have been remarkable advancements in image generation technology.Existing image generation methods mostly generate images by filling pixels,which lacks the stroke-by-stroke drawing process of an artist,resulting in differences in brushstroke details and textures from those found in real artworks.On the contrary,neural painting aims to imitate the stroke-by-stroke drawing process of a human artist with brushstrokes as the basic unit.Most existing neural painting methods simulate real brushstrokes using Bezier curves or stroke templates through affine transformations.However,the lack of texture in Bezier curves and the linear nature of affine transformations indicate that the generated brushstrokes have significant limitations in terms of texture or shape.To better simulate the texture and shape of real brushstrokes,we propose a new curved brushstroke parameter model based on thin plate spline interpolation.By curving and affine-transforming real brushstroke templates in succession,we can generate more realistic and varied brushstroke images.Furthermore,we propose a hierarchical brushstroke optimization method that decomposes the entire image into multiple brushstrokes,from large to small,effectively improving the model's painting ability for both the overall structure and local details of the image.Finally,we extend the proposed method to style transfer,achieving impressive style transfer effects.Qualitative and quantitative experiments demonstrate that our new brushstroke model and optimization method outperform the existing models in both neural painting and stylization tasks.

neural paintingthin plate splinecurved strokelayer optimizationstyle transfer

唐波昊、胡腾、杜瑜桢、易冉、马利庄

展开 >

上海交通大学电子信息与电气工程学院,上海 200240

神经绘画 薄板样条插值 弯曲笔触 层次优化 风格迁移

国家自然科学基金国家自然科学基金国家自然科学基金中国科协青年人才托举工程上海市青年科技英才扬帆计划上海市科委科技创新行动计划上海市人工智能重大专项CCF-腾讯科研基金中国高校基本科研业务费专项资金北京市自然科学基金-海淀原始创新联合基金

6230229762272447721928212022QNRC00122YF1420300215111012002021SHZDZX0102RAGR20220121YG2023QNB17L222117

2024

10.1360/SSI-2023-0194

中国科学F辑
中国科学院,国家自然科学基金委员会

中国科学F辑

CSTPCD北大核心
影响因子：1.438
ISSN：1674-5973
年,卷(期)：2024.54(2)
  • 23