@article{b64802bfcf4c4e55be1c206e3e9a03d8,
title = "Mesh Defiltering via Cascaded Geometry Recovery",
abstract = "This paper addresses the nontraditional but practically meaningful reversibility problem of mesh filtering. This reverse-filtering approach (termed a DeFilter) seeks to recover the geometry of a set of filtered meshes to their artifact-free status. To solve this scenario, we adapt cascaded normal regression (CNR) to understand the commonly used mesh filters and recover automatically the mesh geometry that was lost through various geometric operations. We formulate mesh defiltering by an extreme learning machine (ELM) on the mesh normals at an offline training stage and perform it automatically at a runtime defiltering stage. Specifically, (1) to measure the local geometry of a filtered mesh, we develop a generalized reverse Filtered Facet Normal Descriptor (grFND) in the consistent neighbors; (2) to map the grFNDs to the normals of the ground-truth meshes, we learn a regression function from a set of filtered meshes and their ground-truth counterparts; and (3) at runtime, we reversely filter the normals of a filtered mesh, using the learned regression function for recovering the lost geometry. We evaluate multiple quantitative and qualitative results on synthetic and real data to verify our DeFilter's performance thoroughly. From a practical point of view, our method can recover the lost geometry of denoised meshes without needing to know the exact filter used previously, and can act as a geometry-recovery plugin for most of the state-of-the-art methods of mesh denoising.",
keywords = "CCS Concepts, • Computing methodologies → Shape analysis",
author = "M. Wei and X. Guo and J. Huang and H. Xie and H. Zong and R. Kwan and Wang, {F. L.} and J. Qin",
note = "Funding Information: The authors would like to thank the anonymous reviewers for their constructive suggestions. This work was supported by the grants from the National Natural Science Foundation of China (No. 61502137), the Top‐Up Fund (TFG‐04) and Seed Fund (SFG‐10) for General Research Fund/Early Career Scheme, the Interdisciplinary Research Scheme of the Dean's Research Fund 2018‐19 (FLASS/DRF/IDS‐3), the Departmental Collaborative Research Fund 2019 (MIT/DCRF‐R2/18‐19), the Funding Support to General Research Fund Proposal (RG 39/2019‐2020R), the Internal Research Grant (RG 90/2018‐2019R) of The Education University of Hong Kong, and The Hong Kong Polytechnic University (No. G‐YBZE). Mesh defiltering results on Child (upper) and Boy2 (bottom). From the left column to the right: The upper row shows the ground‐truth Child model, the noisy model (Gaussian noise with σ E = 0.25, the average normal angular difference D n = 27.9), the denoising result (D n = 7.46) by NormalNet, and the refined result (D n = 5.83) by our DeFilter (using the training results of D1) respectively. The bottom row shows the high‐resolution ground‐truth Boy2 model obtained by a laser scanner, the single‐frame noisy mesh (D n = 31.1) by Kinect v1, the denoised result (D n = 10.5) by NormalNet, and the refined result (D n = 8.4) by our DeFilter (using the training results of D2) respectively. Funding Information: The authors would like to thank the anonymous reviewers for their constructive suggestions. This work was supported by the grants from the National Natural Science Foundation of China (No. 61502137), the Top-Up Fund (TFG-04) and Seed Fund (SFG-10) for General Research Fund/Early Career Scheme, the Interdisciplinary Research Scheme of the Dean's Research Fund 2018-19 (FLASS/DRF/IDS-3), the Departmental Collaborative Research Fund 2019 (MIT/DCRF-R2/18-19), the Funding Support to General Research Fund Proposal (RG 39/2019-2020R), the Internal Research Grant (RG 90/2018-2019R) of The Education University of Hong Kong, and The Hong Kong Polytechnic University (No. G-YBZE). Mesh defiltering results on Child (upper) and Boy2 (bottom). From the left column to the right: The upper row shows the ground-truth Child model, the noisy model (Gaussian noise with ?E = 0.25, the average normal angular difference Dn = 27.9), the denoising result (Dn = 7.46) by NormalNet, and the refined result (Dn = 5.83) by our DeFilter (using the training results of D1) respectively. The bottom row shows the high-resolution ground-truth Boy2 model obtained by a laser scanner, the single-frame noisy mesh (Dn = 31.1) by Kinect v1, the denoised result (Dn = 10.5) by NormalNet, and the refined result (Dn = 8.4) by our DeFilter (using the training results of D2) respectively. Publisher Copyright: {\textcopyright} 2019 The Author(s) Computer Graphics Forum {\textcopyright} 2019 The Eurographics Association and John Wiley & Sons Ltd. Published by John Wiley & Sons Ltd.",
year = "2019",
month = oct,
day = "1",
doi = "10.1111/cgf.13863",
language = "English",
volume = "38",
pages = "591--605",
journal = "Computer Graphics Forum",
issn = "0167-7055",
publisher = "Wiley-Blackwell Publishing Ltd",
number = "7",
}