🏙️ A Raster-Based Method for Building Simplification and Texturing

A novel raster-based approach for building simplification and texture preservation in remote sensing imagery, integrating superpixel segmentation, texture matching, and hue adjustment.

Authors: Ruijie Fan, Yilang Shen
Affiliation: School of Geospatial Engineering and Science, Sun Yat-sen University
Published in: Geo-spatial Information Science (2025)
DOI: 10.1080/10095020.2025.2573369

---

🌍 Overview

This repository provides the implementation and experimental data of the paper
“A raster-based method for building simplification considering shape and texture features based on remote sensing images.”

The proposed Shape and Texture-Aware Building Simplification (STABS) method introduces a raster-based approach for simplifying buildings while preserving visual continuity in remote sensing imagery. Unlike traditional vector-based or textureless raster simplification methods, STABS simultaneously considers both shape and texture information to maintain the perceptual consistency of buildings across multiple scales.

---

🚀 Key Contributions

• Texture-aware building simplification:
  Introduces a method that assigns suitable textures from a pre-built texture library by comparing the GLCM and LBP feature similarities between the original building and library textures.

• Hue-adjusted texture matching:
  Applies hue correction to matched textures to ensure color consistency between simplified and original buildings, enhancing cartographic continuity.

• Superpixel-based segmentation:
  Employs the SEEDS algorithm for superpixel segmentation, followed by evaluation using corner ratio (CR) and square ratio (SR) to retain geometrically meaningful building parts.

• Improved raster map generalization:
  The method provides visually stable and structurally faithful simplification results that integrate smoothly back into remote sensing imagery.

---

📊 Experimental Data

• Dataset:
  Building imagery from the Wuhan University Building Dataset, based on aerial data from Land Information New Zealand (LINZ).

• Texture Library:
  Constructed from 50 building textures collected via Maps of Switzerland, processed with the Image Quilting for Texture Synthesis method.

• Evaluation Metrics:
  Quantitative comparison using MSE, PSNR, MS-SSIM, Gradient Similarity (GS), and Edge Preservation Index (EPI) demonstrates that STABS achieves higher similarity to the original imagery than Gaussian filtering and traditional methods.

---

🧩 Repository Contents

├── data/ # Sample dataset and building labels
├── textures/ # Building texture library
├── src/ # Source code for STABS implementation
│ ├── segmentation/ # Superpixel extraction (SEEDS)
│ ├── simplification/ # CR/SR-based building simplification
│ ├── texture_matching/ # GLCM + LBP texture comparison
│ └── hue_adjustment/ # Hue correction and compositing
├── results/ # Experimental outputs and comparisons
└── README.md # Project documentation

---

🧠 Citation

If you use this work, please cite:

Fan, R., & Shen, Y. (2025). A raster-based method for building simplification considering shape and texture features based on remote sensing images.
Geo-spatial Information Science. DOI: 10.1080/10095020.2025.2573369

---

🔮 Future Work

• Incorporate generative models (e.g., GANs) to fill in missing textures and improve realism.
• Extend to 3D-aware simplification, integrating structural line extraction and rooftop geometry.
• Explore deep learning-based end-to-end map generalization to automatically learn simplification strategies.