Abstract: Vector geospatial data is of great value due to the costly processes of acquiring such data. Protecting this data from piracy has become a critical issue in geographic information science, with watermarking emerging as a feasible solution. This paper proposes a blind watermarking technique for protecting vector geospatial data from illegal use, focusing on four key criteria: usability, invisibility, robustness, and blindness. The technique involves a watermark embedding process and an extraction process. In the embedding phase, two feature layers are selected, and key points are identified as embedding positions. The watermark is shuffled and embedded into these layers. The extraction process re-identifies these key points to retrieve the watermark and calculates a similarity degree to verify ownership. Experiments demonstrate that this technique resists attacks from data format changes, random noise, similarity transformations, and data editing.
Introduction
Geospatial data, such as digital maps used in GPS and urban planning, represents the digital "blueprints" of our physical world. Producing high-quality vector geospatial data is an expensive endeavor, often requiring extensive field surveys, satellite imagery processing, and complex cartographic work. However, the digital nature of this data makes it trivial to copy and distribute without authorization.
While digital watermarking—hiding a "signature" within the data—is a known solution, many existing methods for vector data are "non-blind." These require the original, un-watermarked dataset to prove ownership, creating significant logistical hurdles for copyright enforcement. Furthermore, many methods fail to account for the unique geometric properties of lines (roads) and polygons (buildings), leading to watermarks that are easily destroyed by standard map editing or that degrade the map's precision.
Methodology
The proposed approach focuses on four pillars: usability, invisibility, robustness, and blindness. The workflow is divided into two main phases: embedding and extraction.
Watermark Embedding
The embedding process begins with Layer Selection, where the algorithm identifies two critical feature layers (e.g., "Roads" and "Buildings"). This redundancy is vital for the later verification step.
Next, Key Point Selection is performed. Rather than embedding data in random vertices, the system identifies stable geometric points. For linear features like roads, it targets terminals and intersections. For areal features (polygons), it identifies "joint points" or uses a deviation angle algorithm to find vertices that are less likely to be altered during data simplification.
Finally, the watermark (typically a text string) is converted to bits and shuffled to prevent discovery. These bits are embedded into the Least Significant Bits (LSBs) of the coordinates of the identified key points. This ensures that the positional changes are negligible and the data remains usable for professional engineering.
Blind Extraction
Because the key point selection is based on the geometry itself, the extraction process can re-identify those same points in a suspect file without needing the original map. The watermark is extracted from both layers and compared. The "Similarity Degree" between the two extracted signatures determines if the data contains the watermark, with a threshold of 0.70 or higher serving as a reliable confirmation of ownership.
Experimental Evaluation
The researchers evaluated the method using a professional-grade topographic map (1:50,000 scale) provided by the Surveying and Mapping Bureau of Gansu Province, China. The dataset included 14 feature layers in the standard SHAPE file format, specifically targeting the "roads" and "settlements" layers for testing.
Key Results
- Usability & Invisibility: Positional changes to coordinates were well within standard tolerance levels, and visual comparisons showed no perceptible differences between original and watermarked data.
- Robustness: The watermark successfully survived various "attacks," including format conversions, similarity transformations (scaling, rotation, shifting), and data editing (noise addition, point deletion, or data insertion).
- Verification: By embedding the watermark in two different layers, the authors created a self-verifying system where the statistical probability of a false positive is near zero.
Conclusion
This blind watermarking technique offers a robust solution for protecting the intellectual property of geospatial data producers. By leveraging topological stability and LSB embedding, the method ensures that maps remain accurate and usable while providing a reliable means of ownership verification. Future work will explore the technique's effectiveness across more specialized datasets, such as utility networks, and its resistance to more sophisticated desynchronization attacks.
Summary updated