TL;DR
This paper introduces FOLProM, a novel feature-oriented progressive lossless mesh coding technique. It significantly improves the preservation of sharp geometric features at low bitrates, outperforming existing methods in rate-distortion performance, especially for complex and non-manifold meshes.
What Problem Does This Paper Address?
Storing and transmitting high-resolution 3D models is a significant challenge due to their large file sizes. Progressive mesh coding aims to address this by representing a mesh as a sequence of increasingly detailed versions (Levels of Detail or LODs), allowing for efficient transmission and rendering. However, existing methods often struggle to preserve important geometric features (like sharp edges or corners) at low bitrates, leading to visually degraded models. Furthermore, many techniques are limited to manifold meshes, excluding a significant portion of real-world 3D data. This paper tackles the problem of creating a progressive lossless mesh coder that is generic (handles arbitrary topology, including non-manifolds), preserves geometric features effectively, and achieves superior rate-distortion performance, particularly at low bitrates.
Methods
FOLProM employs a feature-oriented approach to progressive lossless mesh coding. The core idea is to generate a hierarchy of LODs through iterative vertex set splitting and bounding volume subdivision. Key novelties include:
- Feature-Oriented Geometry Coding: The entire geometry coding process is optimized to preserve sharp geometric features. This is achieved through:
- Adaptive Hierarchy Control: A distance metric that combines Euclidean distance with normal similarity is used in the Generalized Lloyd Algorithm (GLA) for clustering. This helps to better separate regions with sharp edges.
- Feature-Based Adaptive Quantization: Quantization parameters are adaptively determined based on local surface shape, prioritizing accuracy in more important spatial dimensions.
- Representative Computation: A Gaussian function-based curvature weighting approach is used to select representatives that are close to the geometric center and have high curvature, preserving sharp features.
- Effective Entropy Reduction Techniques:
- Cylindrical Coordinates: Using cylindrical coordinates (p, θ, z) for representing vertex offsets offers advantages over Cartesian coordinates for prediction and adaptive quantization.
- Prediction Schemes: Predictive coding is applied to polar angles and delta-z coordinates, significantly reducing data entropy.
- Delta Coding of Heights: This further contributes to entropy reduction.
- Simple Yet Effective Connectivity Coder: A novel connectivity coder is proposed that handles arbitrary topology (including non-manifolds) and is compatible with the progressive geometry coder. It leverages local vertex distribution and a pivot neighbor prediction scheme.
Datasets
The paper evaluates its method on a variety of standard 3D models, including:
- Star
- David
- Donna (a non-manifold model)
- Dragon
- Feline
- Horse
- Igea
- Maple (a non-manifold triangle soup)
- Rabbit
- Skeleton_hand (a non-manifold model)
- Venus Head
- Fandisk
The paper does not introduce new benchmark datasets but uses these diverse models to demonstrate the robustness and effectiveness of FOLProM across different complexities and topologies.
Key Insights and Ideas
- Feature Preservation is Key at Low Bitrates: The paper strongly emphasizes that preserving sharp geometric features is crucial for visual quality, especially when very few bits are available per vertex. Standard methods often smooth these out.
- Adaptive Metrics Matter: Combining geometric and normal information in distance metrics for clustering (GLA) is vital for correctly identifying and preserving sharp features, preventing the grouping of spatially close but geometrically distinct regions.
- Cylindrical Coordinates Offer Predictive Power: The choice of cylindrical coordinates for representing vertex offsets provides a natural way to exploit spatial relationships and predict values, leading to better compression.
- Generic Handling is Essential: The ability to handle non-manifold meshes and triangle soups is a significant advantage, as many real-world 3D models fall into these categories.
Key Results
- Significant Rate-Distortion (R-D) Gain: FOLProM demonstrates a substantial R-D advantage over prior art, particularly at low bitrates. For example, it shows a significant gain over the OCT coder [PK05a] across various models.
- Superior Visual Quality: Visual comparisons (e.g., in Figure 6) show that FOLProM preserves sharp features much better than other methods like OCT, resulting in visually more appealing intermediate meshes at low bitrates.
- Effective for Non-Manifold Meshes: The method performs well on non-manifold models like Donna and Maple, where other methods might struggle.
- Competitive with State-of-the-Art: Comparisons with wavelet-based (Wavemesh [VP04]) and spectral geometry coders [KG00a] show FOLProM's competitive performance.
- Reduced Hausdorff Distortion: Compared to IPR [VCP09], FOLProM achieves significantly better Hausdorff distortion, indicating better preservation of fine geometric details.
Limitations and Future Work
- Limitations:
- The paper notes that the R-D advantage for triangle soups (like Maple) is not as sharp as for manifold meshes, suggesting that the feature-preservation techniques are most effective on manifold and smooth surfaces.
- The research code for the encoder and decoder is not yet optimized for running efficiency.
- Future Work:
- Explore bottom-up hierarchical construction methods.
- Investigate optimal bit allocation strategies between coding more vertices and more bits per vertex.
- Explore wavelet coding of child offsets using well-designed local shape descriptors as contexts.
- Consider coding original floating-point vertex coordinates with minor adaptations.
Summary updated