A Brushstroke Style Transfer Method from Painting to Map

SUN Yanjie; LAI Zhicheng; ZHOU Chuanlong; WU Mingguang

doi:10.13203/j.whugis20250209

您当前的位置：

首页 >

文章列表页 >

A Brushstroke Style Transfer Method from Painting to Map

更新时间：2026-11-05

- A Brushstroke Style Transfer Method from Painting to Map
- Geomatics and Information Science of Wuhan University Vol. 51, Issue 3, Pages: 463-474(2026)
- 作者机构：
  
  1.南京师范大学地理科学学院，江苏南京，210023
  2.中国电子科技集团公司第二十八研究所，江苏南京，210023
  3.南京师范大学虚拟地理环境教育部重点实验室，江苏南京，210023
  4.江苏省地理信息资源开发与利用协同创新中心，江苏南京，210023
- 作者简介：
- 基金信息：
- DOI：10.13203/j.whugis20250209
  CLC： P208
- Received：15 July 2025，
  
  Published：05 March 2026
- 稿件说明：
移动端阅览
孙彦杰,赖志诚,周传龙，等.从绘画到地图的笔触风格迁移方法[J].武汉大学学报（信息科学版）,2026,51(3):463-474.

SUN Yanjie,LAI Zhicheng,ZHOU Chuanlong,et al.A Brushstroke Style Transfer Method from Painting to Map[J].Geomatics and Information Science of Wuhan University,2026,51(3):463-474.
孙彦杰,赖志诚,周传龙，等.从绘画到地图的笔触风格迁移方法[J].武汉大学学报（信息科学版）,2026,51(3):463-474. DOI：10.13203/j.whugis20250209

SUN Yanjie,LAI Zhicheng,ZHOU Chuanlong,et al.A Brushstroke Style Transfer Method from Painting to Map[J].Geomatics and Information Science of Wuhan University,2026,51(3):463-474. DOI：10.13203/j.whugis20250209

摘要

地图风格迁移是指将颜

色、纹理、笔触等视觉元素从绘画等作品转移到地图上的过程，好的风格迁移能够提升个性、创意地图制作的效率和质量。在绘画中，笔触包含了线条的粗细、纹理及笔画的方向等丰富的线状图形信息，是绘画风格的关键特征，也是艺术家表达情感的重要视觉元素。从绘画到地图的笔触风格迁移，有利于提升地图的视觉表现力，加强地图的形式美与情感表达效果，在城市印象地图、旅游地图等类型的泛地图制作方面有着迫切的技术需求。当前的地图风格迁移主要集中在颜色、纹理等方面，缺少针对笔触的地图风格迁移研究，通用的绘画笔触风格迁移方法通常忽略了地图空间关系保持与笔触风格表达之间的平衡，容易导致损失地图的信息负载及传输功能。为解决这一问题，基于扩散模型提出一种模拟绘画过程的从绘画到地图的笔触风格迁移方法。该方法借鉴人类绘画从结构框架到笔触填充的过程，并结合地图符号的绘制策略，区分符号几何维度和图形-背景关系。通过将笔触风格迁移视为扩散模型下的渐进式笔触匹配过程，能够在保留地图空间关系准确性的基础上实现笔触风格的自适应迁移。实验在多个指标上进行了评估，包括结构相似性、内容保真度、风格相似性与符号清晰性，结果显示，所提方法在多个方面均表现出较优的迁移效果（

0.05），有望丰富地图风格迁移的方法体系，为个性创意制图提供新的技术途径。

Abstract

Objectives

Map style transfer involves the integration of visual elements such as color

texture

and brushstrokes from artistic works into cartographic design. Among these

brushstrokes convey essential stylistic features and emotional tones through attributes like line thickness

directionality

and texture. Incorporating brushstroke styles into maps enhances their expressive capacity and aesthetic appeal

especially in personalized and thematic cartography such as urban impression maps and cultural or tourist maps. Existing research has largely focused on color and texture transfer

with limited attention to brushstroke-specific migration in a cartographic context.

Methods

To address the lack of brushstroke-focused transfer methods to maps

a diffusion model-based approach is proposed that simulates the human painting process. First

control points are identified for various map elements: for point features

the coordinates are directly used; for line features

all vertices forming the line are selected; for polygonal and background features

contour characteristics are extracted using a self-attention mechanism and transformed into contour control points. These control points from point

line

area

and background elements are then unified to form a complete map framework control point set. Next

Bézier curves are generated from these control points and used as initial strokes for iterative optimization of the map framework. During optimization

a differentiable rasterizer converts Bézier curves into differentiable raster images

enabling the generation process to be integrated into deep learning models for further refinement. The optimization model takes the Bézier control points and opacity as inputs

using alpha blending to simulate stroke intensity and overlap effects

and applies an extended score distillation sampling to minimize differences between the rasterized image and the input map. Parameters are iteratively updated until convergence to produce an accurate brushstroke-based map framework. Building on this

inspired by the What‐Where dual-channel mechanism of human vision

specific loss functions are designed for different stroke types to jointly constrain the diffusion model's optimization in terms of spatial relationship and stylistic features. The Where channel ensures spatial consistency between the generated and input maps

while the What channel evaluates stylistic similarity between the generated map and the reference painting. Finally

the diffusion model progressively matches and stylizes the map brushstrokes

achieving adaptive brushstroke transfer while preserving figure‐ground relationships and symbol geometric precision.

Results

The proposed method was evaluated against representative brushstroke transfer techniques through both quantitative and qualitative analyses across four key metrics: structural similarity

content fidelity

style consistency

and symbol clarity. For structural similarity

the structural similarity index measure (SSIM) was employed to quantitatively assess the generated maps

where the proposed method achieved an average SSIM score of 0.876 across two test maps

substantially higher than the comparison method's 0.333

indicating superior preservation of the spatial structure of map content. Regarding content fidelity

the content fidelity (CF) metric was us

ed to evaluate the consistency of cartographic information before and after style transfer

with the proposed method achieving an average CF score of 0.911 compared to 0.551 for the comparison method

demonstrating higher precision in content restoration and information retention. To further examine perceptual differences in style transfer performance

a user study was conducted employing a 7-point Likert scale to evaluate style consistency and symbol clarity. In terms of style consistency

maps generated by the proposed method consistently received higher ratings than those of the comparison method

with mean improvements exceeding 1 point and all differences reaching statistical significance (

0.05). In terms of symbol clarity

the proposed method outperformed the comparison method across all test cases

achieving average improvements of over 3 points

with all results statistically significant. Overall

experimental results indicate that the proposed method outperforms the representative brushstroke transfer method across all four evaluation metrics

with statistically significant improvements (

0.05)

particularly in maintaining the legibility and semantic accuracy of map elements under stylization.

Conclusions

The proposed method employs diffusion-based modeling to guide the progressive transfer of brushstroke styles from paintings to maps

balancing the preservation of spatial relationship and symbol semantics. This approach also provides a feasible solution for integrating brushstroke styles into maps

facilitating further exploration of the coordination of style and structure in cartographic design. Nevertheless

the method still has room for improvement regarding map information load

transmission efficiency

and brushstroke adaptability. Future research should further examine the impact of artistic style on map readability and information communication

and investigate the adaptability and transferability of reference artworks

to achieve an optimal balance between artistic expression and cartographic functionality.

关键词

Keywords

references

GATYS L A ， ECKER A S ， BETHGE M . Image Style Transfer Using Convolutional Neural Networks ［C］// IEEE Conference on Computer Vision and Pattern Recognition （CVPR）， Las Vegas， NV， USA， 2016 .

吴明光，孙彦杰，闾国年 . 地图风格迁移：概念、综述与展望［J］. 武汉大学学报（信息科学版）， 2022 ， 47 （ 12 ）： 2069 - 2084 .

WU Mingguang ， SUN Yanjie ， LÜ Guonian . Cartographic Style Transfer： Idea， Review and Envision ［J］. Geomatics and Information Science of Wuhan University ， 2022 ， 47 （ 12 ）： 2069 - 2084 .

任福，翁杰，王昭，等 . 关于智能地图制图的几点思考［J］. 武汉大学学报（信息科学版）， 2022 ， 47 （ 12 ）： 2064 - 2068 .

REN Fu ， WENG Jie ， WANG Zhao ， et al . Some Thoughts on Smart Cartography ［J］. Geomatics and Information Science of Wuhan University ， 2022 ， 47 （ 12 ）： 2064 - 2068 .

艾廷华 . 深度学习赋能地图制图的若干思考［J］. 测绘学报， 2021 ， 50 （ 9 ）： 1170 - 1182 .

AI Tinghua . Some Thoughts on Deep Learning Enabling Cartography ［J］. Acta Geodaetica et Cartographica Sinica ， 2021 ， 50 （ 9 ）： 1170 - 1182 .

孙群，任福，龙毅，等 . 中国地图制图学与地理信息工程进展报告（2019—2023）［J］. 测绘学报， 2024 ， 53 （ 3 ）： 399 - 412 .

SUN Qun ， REN Fu ， LONG Yi ， et al . The Progress and Trend of Cartography and Geographic Information Engineering in China （2019—2023）［J］. Acta Geodaetica et Cartographica Sinica ， 2024 ， 53 （ 3 ）： 399 - 412 .

郭仁忠，陈业滨，赵志刚，等 . 泛地图学理论研究框架［J］. 测绘地理信息， 2021 ， 46 （ 01 ）： 9 - 15 .

GUO Renzhong ， CHEN Yebin ， ZHAO Zhigang ， et al . A Theoretical Framework for the Study of Pan-Maps ［J］. Journal of Geomatics ， 2021 ， 46 （ 11 ）： 9 - 15 .

郭仁忠，陈业滨，赵志刚，等 . ICT时代地图的科学概念及表达框架［J］. 武汉大学学报（信息科学版）， 2022 ， 47 （ 12 ）： 1978 - 1987 .

GUO Renzhong ， CHEN Yebin ， ZHAO Zhigang ， et al . Scientific Concept and Representation Framework of Maps in the ICT Era ［J］. Geomatics and Information Science of Wuhan University ， 2022 ， 47 （ 12 ）： 1978 - 1987 .

郭仁忠，陈业滨，应申，等 . 三元空间下的泛地图可视化维度［J］. 武汉大学学报（信息科学版）， 2018 ， 43 （ 11 ）： 1603 - 1610 .

GUO Renzhong ， CHEN Yebin ， YING Shen ， et al . Geographic Visualization of Pan-Map with the Context of Ternary Spaces ［J］. Geomatics and Information Science of Wuhan University ， 2018 ， 43 （ 11 ）： 1603 - 1610 .

闫浩文，张黎明，杜萍，等 . 自媒体时代的地图：微地图［J］. 测绘科学技术学报， 2016 ， 33 （ 5 ）： 520 - 523 .

YAN Haowen ， ZHANG Liming ， DU Ping ， et al . We-Map： A New Type of Map in the Era of We Media ［J］. Journal of Geomatics Science and Technology ， 2016 ， 33 （ 5 ）： 520 - 523 .

WU M G ， SUN Y J ， LI Y Q . Adaptive Transfer of Color from Images to Maps and Visualizations ［J］. Cartography and Geographic Information Science ， 2022 ， 49 （ 4 ）： 289 - 312 .

JENNY B ， HEITZLER M ， SINGH D ， et al . Cartographic Relief Shading with Neural Networks ［J］. IEEE Transactions on Visualization and Computer Graphics ， 2021 ， 27 （ 2 ）： 1225 - 1235 .

SUN Y J ， WU M G . Flexible Style Transfer from Remote Sensing Images to Maps ［J］. International Journal of Applied Earth Observation and Geoinformation ， 2025 ， 139 ： 104566 .

HERTZMANN A . Painterly Rendering with Curved Brush Strokes of Multiple Sizes ［C］// The 25th Annual Conference on Computer Graphics and Interactive Techniques ， New York，NY，USA， 1998 .

YAO Y ， REN J Q ， XIE X S ， et al . Attention-Aware Multi-stroke Style Transfer ［C］// IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）， Long Beach， CA， USA， 2020 .

LIU X C ， WU Y C ， HALL P . Painterly Style Transfer with Learned Brush Strokes ［J］. IEEE Transactions on Visualization and Computer Graphics ， 2024 ， 30 （ 9 ）： 6309 - 6320 .

MACKANESS W A . An Algorithm for Conflict Identification and Feature Displacement in Automated Map Generalization ［J］. Cartography and Geographic Information Systems ， 1994 ， 21 （ 4 ）： 219 - 232 .

BOUTS Q W ， DWYER T ， DYKES J ， et al . Visual Encoding of Dissimilarity Data via Topology-Preserving Map Deformation ［J］. IEEE Transactions on Visualization and Computer Graphics ， 2016 ， 22 （ 9 ）： 2200 - 2213 .

JING Y C ， LIU Y ， YANG Y Z ， et al . Stroke Controllable Fast Style Transfer with Adaptive Receptive Fields ［M］// Computer Vision — ECCV 2018 . Cham ： Springer International Publishing ， 2018 ： 244 - 260 .

KANG Y H ， GAO S ， ROTH R E . Transferring Multiscale Map Styles Using Generative Adversarial Networks ［J］. International Journal of Cartography ， 2019 ， 5 （ 2/3 ）： 115 - 141 .

HERTZMANN A . A Survey of Stroke-Based Rendering ［J］. IEEE Computer Graphics and Applications ， 2003 ， 23 （ 4 ）： 70 - 81 .

XING X M ， WANG C ， ZHOU H T ， et al . DiffSketcher： Text Guided Vector Sketch Synthesis Through Latent Diffusion Models ［C］// The 37th International Conference on Neural Information Processing Systems ， New Orleans， Louisiana， USA， 2023 .

LI T M ， LUKÁČ M ， GHARBI M ， et al . Differentiable Vector Graphics Rasterization for Editing and Learning ［J］. ACM Transactions on Graphics ， 2020 ， 39 （ 6 ）： 1 - 15 .

SAHARIA C ， CHAN W ， SAXENA S ， et al . Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding ［C］// The 36th International Conference on Neural Information Processing Systems ， New Orleans， Louisiana， USA，2022 .

MO H R ， SIMO-SERRA E ， GAO C Y ， et al . General Virtual Sketching Framework for Vector Line Art ［J］. ACM Transactions on Graphics ， 2021 ， 40 （ 4 ）： 1 - 14 .

DHARIWAL P ， NICHOL A . Diffusion Models Beat GANs on Image Synthesis ［C］// The 35th International Conference on Neural Information Processing Systems ， New York，NY，United States，2021 .

ZHANG X ， YUNIS D ， MAIRE M . Deciphering ‘What’ and ‘Where’ Visual Pathways from Spectral Clustering of Layer-Distributed Neural Representations ［C］// IEEE/CVF Conference on Computer Vision and Pattern Recognition （CVPR）， Seattle， WA， USA， 2024 .

阮国庆，杨守威，徐琳，等 . 基于提示的视觉语言基础模型综述［J］. 指挥信息系统与技术， 2024 ， 15 （ 6 ）： 12 - 20 .

RUAN Guoqing ， YANG Shouwei ， XU Lin ， et al . Review of Prompt-Based Vision Language Foundation Models ［J］. Command Information System and Technology ， 2024 ， 15 （ 6 ）： 12 - 20 .

IOANNOU E ， MADDOCK S . Evaluation in Neural Style Transfer： A Review ［J］. Computer Graphics Forum ， 2024 ， 43 （ 6 ）： e15165 .

WANG Z ， BOVIK A C ， SHEIKH H R ， et al . Image Quality Assessment： From Error Visibility to Structural Similarity ［J］. IEEE Transactions on Image Processing ， 2004 ， 13 （ 4 ）： 600 - 612 .

WANG Z Z ， ZHAO L ， CHEN H B ， et al . Evaluate and Improve the Quality of Neural Style Transfer ［J］. Computer Vision and Image Understanding ， 2021 ， 207 ： 103203 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Cartographic Style Transfer： Idea， Review and Envision

Advances in Artificial Intelligence-Driven Modeling， Prediction， and Applications of GNSS Atmospheric Effects

From Task-Specific Models to General-Purpose Model-Driven Paradigm Shift： A Review and Future Directions of GeoAI Research

Current Status and Trends of Artificial Intelligence Geodesy

Constructing an Urban Waterlogging Depth Grade Assessment Model Using Images Captured by the Public

Related Author

SUN Yanjie

LÜ Guonian

ZHOU Yaozong

LOU Yidong

BI Haohang

LIU Mengjie

ZHANG Weixing

LUO Xiaomin

Related Institution

Institute of Geodesy and Photogrammetry, Eidgenössische Technische Hochschule Zürich, Zürich

GNSS Research Center, Wuhan University

Hubei Luojia Laboratory

School of Geography and Information Engineering, China University of Geosciences

School of Surveying and Mapping, Wuhan University

AI问答

⁰