Abstract
3D digitization is becoming the norm in the preservation of tangible cultural properties be it movable or immovable. In the case of mobile cultural objects, 3D models are normally acquired through non-destructive methods such as photogrammetry or 3D scanning using structured light or laser. The continuing improvements in scanning equipment which result in higher accuracy and resolution can be translated to mean higher level of 3D data acquired. At times, the quantity of 3D data of a single model can get too large and beyond what a normal workstation can process. This is where optimization comes in. However, the level of optimization will vary to suit the final purpose or usage of the 3D model. During optimization, the resulting optimized 3D model is bound to suffer from some level of data loss when compared to the original 3D model acquired directly from laser scanning. This paper discusses a practical method devised by Amber Digital Solutions in maintaining a high level of accuracy of the final optimized model for national archival purposes which requires the highest form of quality.
Introduction
In the world of digital heritage preservation, 3D scanning via photogrammetry, structured light, or laser has become the gold standard for documenting cultural artifacts. However, these high-fidelity scans generate "point clouds" so dense that they frequently exceed the processing capabilities of standard workstations. To make these models usable for researchers, curators, or public dissemination, they must undergo a process of optimization—reducing polygon counts and smoothing surfaces to create manageable file sizes.
The challenge lies in the fact that standard optimization algorithms, often designed for industrial components, can fail when applied to cultural relics. While industrial tools prioritize dimensional precision, cultural artifacts require the preservation of "meticulous craftsmanship"—the subtle textures, sharp edges, and intricate adornments created by human hands. Blindly applying optimization often leads to "blunted" edges, the loss of facial features, or the accidental "patching" of intentional grooves that software mistakes for holes in the data. This paper introduces a practical quality control workflow and a software tool called "Amber Form" that quantifies data loss by comparing optimized models against original high-density scans.
The 3D Scanning Pipeline
The authors outline a standard 3D scanning pipeline that moves from initial setup to the final rendered model. This workflow is essential for ensuring that the data captured is both accurate and comprehensive.
As shown in Figure 1, the process begins with the calibration of the 3D scanner, followed by data acquisition which includes both point cloud generation and high-resolution photography. After global registration aligns the various scans, the model enters the optimization phase. It is at this stage that the risk of data loss is highest, necessitating a robust Quality Control (QC) mechanism.
Methods: Quality Control via Amber Form
The core of the proposed approach is a comparative analysis implemented in the proprietary software, Amber Form. This method allows practitioners to move beyond visual guesswork and into quantitative assessment. The workflow functions as follows:
- Reference vs. Target: The original, high-density raw scan is loaded as the "Reference Model," while the decimated, optimized mesh is loaded as the "Target Model."
- Point-to-Point Matching: The software calculates the distance between every point in the optimized model and its nearest neighbor in the original scan.
- Thresholding: Users set a specific distance threshold (e.g., 0.5mm). If a point in the optimized model deviates from the original by more than this distance, it is flagged as an error.
- Color-Coded Visualization: To make the data actionable, the software color-codes the model: Blue indicates a perfect match, Green/Yellow indicates acceptable deviation (1-3%), and Red indicates points that exceed the threshold.
Key Insights and Results
Through real-world case studies, such as a wooden Buddha statue from the National Museum of China and an ancient Chinese ink pad, the authors identified several critical issues in standard optimization:
- The "Noise" Trap: Algorithms often mistake sparse point clouds at sharp edges for noise and delete them, resulting in unnatural "blunted" edges.
- Inaccurate Data Addition: Software may perceive deep, undulating grooves as "holes" and attempt to "patch" them, adding inaccurate geometry to the model.
Quantitative testing demonstrated how thresholding affects the perceived accuracy of a model. As the accuracy requirements become more stringent, the match rate drops, providing a clear metric for quality (Table 1).
| Test | Threshold Value | Point Clouds Matched / Total | Match Rate |
|---|---|---|---|
| 1 | 1.0mm | 6907 / 7090 | 97.42% |
| 2 | 0.5mm | 6836 / 7090 | 96.42% |
| 3 | 0.3mm | 6425 / 7090 | 90.62% |
Conclusion
By using a "Match Rate" based on objective thresholds, curators can finally set rigorous standards for digital archives (e.g., requiring a 95% match rate at a 0.5mm threshold). While the current process identifies errors, future work aims to integrate these QC metrics directly into decimation algorithms, allowing software to automatically "protect" high-detail areas during the optimization process.
Summary updated