Template Matching and Simplification Method for Building Features Based on Shape Cognition

Xiang Zhang (张翔), Xiongfeng Yan, Tinghua Ai, Xiang ZhangIsprs Int. J. geo inf 2017 · DOI

TL;DR

This paper introduces a novel method for simplifying building features in topographic maps by using shape cognition and template matching. By abstracting building shapes into symbolic templates, the method aims to preserve essential structural characteristics and improve map readability, outperforming traditional geometric simplification techniques.

What Problem Does This Paper Address?

Topographic maps are crucial for various applications like urban planning and navigation. However, at different scales, building features need to be simplified to maintain map legibility without losing their essential information. Traditional simplification methods often focus on geometric details, which can lead to a loss of recognizable shapes and patterns. This paper addresses the challenge of simplifying building polygons in a way that captures their inherent shape cognition – how humans perceive and interpret them – and preserves their symbolic meaning and community structure, especially for well-known landmarks and regional patterns.

Methods

The proposed method combines shape cognition with template matching. It involves two main steps:

  1. Shape Measurement: The "turning function" is employed as a shape descriptor. This function represents a polygon's shape as a piecewise function of tangent angle versus arc length. It's chosen for its invariance to translation and scaling, and its ability to handle mirror images, making it suitable for comparing building shapes to templates. The similarity between a building and a template is measured using the L2 distance between their turning functions.
  2. Template Matching: Once a candidate template is identified as the "most similar" to the original building, a 2D Helmert Transformation Model is used to adjust its position, orientation, and scale to match the original building. This is achieved through a least-squares adjustment to minimize the error between the transformed template and the original building's outline. The paper also mentions using a dynamic programming algorithm to find correspondences between building and template outlines.

The paper categorizes templates into four types: Simple Shape (basic geometrics like rectangles), Symbolic Shape (using letters or familiar objects), Composite Shape (combining components for complex structures), and Automatic Extracted Shape (generated through information extraction and classification).

Datasets

The primary dataset used for experiments was a large-scale 1:2000 topographic map of Guangzhou, China, covering an area of approximately 1.2 km x 1.3 km with 625 building polygons. This dataset represents typical residential communities. An additional experiment was conducted using a 1:50,000 scale topographic map from Sichuan, China, to test the method on scattered houses in a medium-small scale context.

Key Insights and Ideas

  • Shape Cognition as a Foundation: The core idea is that simplifying buildings should align with how humans perceive and interpret shapes, not just their geometric intricacies. This means preserving symbolic meaning and recognizable patterns.
  • Turning Function's Suitability: The turning function is highlighted as a robust shape descriptor for artificial objects like buildings, offering invariance to transformations and good performance in similarity measurements.
  • Template Hierarchy: The paper suggests that templates can have a hierarchical structure, where different templates might represent the same symbolic meaning but with varying levels of detail or geometric precision, impacting simplification accuracy.
  • Regularity Over Absolute Accuracy: For map generalization, achieving regular and uniform shapes through template matching can be more beneficial for map readability than strictly adhering to precise geometric accuracy, especially when compared to traditional methods.

Key Results

  • The proposed method successfully simplified a majority of buildings in the Guangzhou dataset, preserving their main structure and geometric accuracy.
  • Qualitative evaluations showed that similar buildings were often simplified by the same template, resulting in more regular and uniform shapes.
  • The method demonstrated effectiveness in simplifying shapes that resemble letters (using symbolic templates) and special shapes (using custom templates).
  • Quantitative evaluation using surface distance showed that the proposed method achieved a high satisfaction rate (94.2% with surface distances between 0.05 and 0.25).
  • Comparison with existing methods (MBR-based and Geometry-based) indicated that the proposed method is more stable (lower standard deviation in surface distance) and generally performs better, especially for non-rectangular shapes. The MBR-based method showed a slight advantage in mean surface distance but was less stable.

Limitations and Future Work

  • Template Library Completeness: The performance of the method is significantly impacted by the completeness and diversity of the template library. Some complex buildings could not be simplified due to the lack of a suitable template.
  • Matching Errors: Errors can occur due to an inability to find a suitable template for very complex shapes, incorrect template matching (false positives), or issues with the transformation step.
  • Scaling Impact: The scaling of templates is crucial and needs careful consideration, balancing computational cost with generalization requirements.
  • Future Work: The authors suggest enriching the template library to include more complex shapes (e.g., buildings with holes) and incorporating other data sources like street networks to avoid potential conflicts and improve simplification accuracy. Further research into shape representation methods and precise orientation definition is also recommended.

Summary updated

BibTeX
@article{2017template,
  author = {Xiang Zhang (张翔) and Xiongfeng Yan and Tinghua Ai and Xiang Zhang},
  title = {{Template Matching and Simplification Method for Building Features Based on Shape Cognition}},
  journal = {Isprs Int. J. geo inf},
  year = {2017},
  doi = {10.3390/ijgi6080250},
}

Loading related content...

Figure