空间频域特征编码下的三维高频表面重建

doi:10.3778/j.issn.1002-8331.2406-0106

摘要/Abstract

摘要： 针对现有神经隐式表面重建算法在复杂物体表面高频细节获取、三维体纹理表征以及重建精度方面存在的不足，提出了一种针对高频细节提取的空间频域特征编码网络。该网络利用空间特征三平面有效建模三维点相互依赖关系，并采用基于快速傅里叶变换的高频特征编码模块来增强其特征提取过程的深层频率变化，使生成的物体表面具有更多高频几何细节。为保持重建过程稳定性，构建了一种轻量级MLP解码器来提高表面重建的保真度并抑制空间频域编码过程中产生的噪声。该方法在DTU和NeRF Synthetic 360°数据集中的六个场景下进行测试，与其他算法进行倒角距离和峰值信噪比指标的定量和定性评估，并针对复杂物体前景细节定量评估。实验结果表明，高频特征编码模块和轻量级MLP解码器的引入显著提高了复杂物体的重建精度，重建出的三维几何与纹理表面恢复了更多精细的几何细节。

关键词: 三维表面重建, 快速傅里叶变换, 符号距离函数, 多层感知机, 神经辐射场

Abstract: Aiming at the deficiencies of existing neural implicit surface reconstruction algorithms in capturing high-frequency details on complex object surfaces, representing 3D textures, and reconstruction accuracy, this paper proposes a spatial frequency domain feature encoding network specifically designed for extracting high-frequency details. This network utilizes spatial feature triplanes to effectively model the interdependencies among 3D points and employs a high-frequency feature encoding module based on the fast Fourier transform (FFT) to enhance deep frequency variations during feature extraction, resulting in object surfaces with richer high-frequency geometric details. To maintain stability during the reconstruction process, a lightweight MLP is constructed to improve the fidelity of surface reconstruction and suppress noise generated during spatial frequency domain encoding. The proposed method is tested on six scenes from the DTU and NeRF Synthetic 360° datasets. This paper compares it against other algorithms using chamfer distance (CD) and peak signal-to-noise ratio (PSNR) metrics, and quantitatively evaluates the foreground details of complex objects. Experimental results demonstrate that the introduction of the high-frequency feature encoding module and the lightweight MLP decoder significantly enhances the reconstruction accuracy of complex objects, with the reconstructed 3D geometry and texture surfaces recovering finer geometric details.

Key words: three-dimensional surface reconstruction, fast Fourier transform, sign distance function, multi-layer perceptron, neural radiance field

魏东, 张静恬, 白宜凡, 孙赫. 空间频域特征编码下的三维高频表面重建[J]. 计算机工程与应用, 2025, 61(18): 241-251.

WEI Dong, ZHANG Jingtian, BAI Yifan, SUN He. Three-Dimensional High-Frequency Surface Reconstruction Under Spatial Frequency Domain Feature Encoding[J]. Computer Engineering and Applications, 2025, 61(18): 241-251.

参考文献

[1] MILDENHALL B, SRINIVASAN P P, TANCIK M, et al. NeRF: representing scenes as neural radiance fields for view synthesis[C]//Proceedings of the 16th European Conference on Computer Vision. Cham: Springer, 2020: 405-421.
[2] 叶子鹏, 夏雯宇, 孙志尧, 等. 从传统渲染到可微渲染: 基本原理、方法和应用[J]. 中国科学(信息科学), 2021, 51(7): 1043-1067.
YE Z P, XIA W Y, SUN Z Y, et al. From traditional rendering to differentiable rendering: theories, methods and applications[J]. Scientia Sinica (Informationis), 2021, 51(7): 1043-1067.
[3] 许威威, 周漾, 吴鸿智, 等. 可微绘制技术研究进展[J]. 中国图象图形学报, 2021, 26(6): 1521-1535.
XU W W, ZHOU Y, WU H Z, et al. Differential rendering: a survey[J]. Journal of Image and Graphics, 2021, 26(6): 1521-1535.
[4] TEWARI A, THIES J, MILDENHALL B, et al. Advances in neural rendering[J]. Journal of the European Association for Computer Graphics, 2022, 41(2):703-735.
[5] WANG P, LIU L, LIU Y, et al. NeuS: learning neural implicit surfaces by volume rendering for multi-view reconstruction[C]//Proceedings of the 35th International Conference on Neural Information Processing Systems, 2021:27171-27183.
[6] YARIV L, GU J, KASTEN Y, et al. Volume rendering of neural implicit surfaces[C]//Advances in Neural Information Processing Systems, 2021:4805-4815.
[7] YU Z, PENG S Y, NIEMEYER M, et al. MonoSDF: exploring monocular geometric cues for neural implicit surface reconstruction[C]//Advances in Neural Information Processing Systems, 2022:25018-25032.
[8] YE S, HU Y, LIN M, et al. Indoor scene reconstruction with fine-grained details using hybrid representation and normal prior enhancement [J]. arXiv:2309.07640, 2023.
[9] CHEN A P, XU Z X, GEIGER A, et al. TensoRF: tensorial radiance fields[C]//Proceedings of the European Conference on Computer Vision. Cham: Springer, 2022: 333-350.
[10] CHAN E R, LIN C Z, CHAN M A, et al. Efficient geometry-aware 3D generative adversarial networks[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 16102-16112.
[11] BAHAT Y, ZHANG Y, SOMMERHOFF H, et al. Neural volume super-resolution[J]. arXiv:2212.04666, 2022.
[12] LI M, ZHOU P, LIU J W, et al. Instant3D: instant text-to-3D generation[J]. International Journal of Computer Vision, 2024, 132(10): 4456-4472.
[13] SHUE J R, CHAN E R, PO R, et al. 3D neural field generation using triplane diffusion[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 20875-20886.
[14] JANSEN J, BAVOIL L. Fourier opacity mapping[C]//Proceedings of the 2010 ACM SIGGRAPH Symposium on Interactive 3D Graphics and Games. New York: ACM, 2010: 165-172.
[15] TEWARI A, FRIED O, THIES J, et al. State of the art on neural rendering[J]. Computer Graphics Forum, 2020, 39(2): 701-727.
[16] LINDELL D B, VAN VEEN D, PARK J J, et al. BACON: band-limited coordinate networks for multiscale scene representation[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 16231-16241.
[17] WANG Y Q, SKOROKHODOV I, WONKA P. HF-NeuS: improved surface reconstruction using high-frequency details[C]//Advances in Neural Information Processing Systems, 2022:1966-1978.
[18] WANG Y Q, SKOROKHODOV I, WONKA P. PET-NeuS: positional encoding tri-planes for neural surfaces[C]//Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2023: 12598-12607.
[19] ZHAO X D, ZHANG M M, TAO R, et al. Fractional Fourier image transformer for multimodal remote sensing data classification[J]. IEEE Transactions on Neural Networks and Learning Systems, 2024, 35(2): 2314-2326.
[20] FRIGO M, JOHNSON S G. The design and implementation of FFTW3[J]. Proceedings of the IEEE, 2005, 93(2): 216-231.
[21] JAMES W COOLEY J W T. An algorithm for the machine calculation of complex Fourier series[J]. Mathematics of Computation, 1965, 19(90): 297-301.
[22] CHATTERJEE S, ZIELINSKI P. On the generalization mystery in deep learning[J]. arXiv:2203.10036, 2023.
[23] ATZMON M, LIPMAN Y. SAL: sign agnostic learning of shapes from raw data[C]//Proceedings of the 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 2562-2571.
[24] KANG G, LEE Y, PARK E. CodecNeRF: toward fast encoding and decoding, compact, and high-quality novel-view synthesis[J]. arXiv:2404.04913, 2024.
[25] JENSEN R, DAHL A, VOGIATZIS G, et al. Large scale multi?view stereopsis evaluation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2014:406-413.
[26] MILDENHALL B, SRINIVASAN P P, TANCIK M, et al. NeRF: representing scenes as neural radiance fields for view synthesis[C]//Proceedings of the European Conference on Computer Vision, 2021: 99-106.
[27] YARIV L, KASTEN Y, MORAN D, et al. Multiview neural surface reconstruction by disentangling geometry and appearance[C]//Proceedings of 2020 IEEE/ CVF Conference on Computer Vision and Pattern Recognition, 2020: 2492-2502.
[28] ZHUANG Y Y, ZHANG Q, FENG Y, et al. Anti-aliased neural implicit surfaces with encoding level of detail[C]//Proceedings of the SIGGRAPH Asia 2023 Conference Papers. New York: ACM, 2023: 1-10.