Mesh Denoising Using Filtering Coefficients Jointly Aware of Noise and Geometry
Abstract: Mesh denoising is a fundamental task in geometry processing, and recent studies have demonstrated the remarkable superiority of deep learning-based methods in this field. However, existing works commonly rely on neural networks without explicit designs for noise and geometry which are actually fundamental factors in mesh denoising. In this paper, by jointly considering noise intensity and geometric characteristics, a novel Filtering Coefficient Learner (FCL for short) for mesh denoising is developed, which delicately generates coefficients to filter face normals. Specifically, FCL produces filtering coefficients consisting of a noise-aware component and a geometry-aware component. The first component is inversely proportional to the noise intensity of each face, resulting in smaller coefficients for faces with stronger noise. For the effective assessment of the noise intensity, a noise intensity estimation module is designed, which predicts the angle between paired noisy-clean normals based on a mean filtering angle. The second component is derived based on two types of geometric features, namely the category feature and face-wise features. The category feature provides a global description of the input patch, while the face-wise features complement the perception of local textures. Extensive experiments have validated the superior performance of FCL over state-of-the-art works in both noise removal and feature preservation.
Relevance To Conference: With the advancement of technologies like realistic 3D rendering and digital twins, 3D data has gained prominence as a crucial modality in multimodal processing. However, the accuracy limitations of scanning equipment inevitably introduce noise into 3D data, severely impacting the performance of multimodal processing tasks. In this context, this work addresses the challenges associated with denoising 3D meshes by leveraging both noise and geometric information, thereby contributing to the field of multimodal processing. The proposed method enhances the quality and reliability of 3D data, thereby benefiting the overall multimodal processing pipeline. Furthermore, in applications involving the creation of digital twins, which entail generating virtual replicas of real-world objects or environments, denoising 3D data becomes crucial for achieving high-fidelity representations. By effectively eliminating noise and preserving geometric details, the proposed method significantly contributes to generating cleaner and more accurate digital twins, ultimately enhancing their reliability and utility in multimodal processing tasks.
Primary Subject Area: [Content] Vision and Language
Secondary Subject Area: [Content] Media Interpretation
Submission Number: 2718
Loading