FD-SCU: Frequency Decomposition-Based Spectrum Collaborative Upsampling for Point Cloud Color Attribute.

Area of Science:

• Computer Vision
• Computer Graphics
• Signal Processing

Background:

• Current point cloud color upsampling methods primarily use interpolation in spatial or feature domains.
• These methods often neglect the potential of the frequency domain for capturing color correlations within local point sets.

Purpose of the Study:

• To propose a novel spectrum collaborative strategy for enhancing point cloud color reconstruction.
• To address the limitations of existing methods by incorporating frequency domain analysis.

Main Methods:

• Voxelization of low-resolution (LR) color point clouds into voxel blocks (VBs).
• A virtual filling strategy to ensure color information occupies VBs.
• Discrete Cosine Transform (DCT) for frequency decomposition into high-frequency (HF) and low-frequency (LF) components.
• Multi-scale cross-domain feature extraction using LF coefficients and LR color data.
• Gaussian perturbation-based feature expansion for initial upsampled color features.
• High-frequency-guided residual refinement using HF coefficients for final high-fidelity output.

Main Results:

• The proposed spectrum collaborative strategy effectively utilizes frequency domain information for color upsampling.
• Experimental results show superior performance compared to state-of-the-art point cloud color upsampling methods.
• The method successfully produces high-fidelity upsampled color point clouds.

Conclusions:

• The frequency decomposition approach offers a significant improvement over traditional interpolation-based methods.
• The proposed method provides a robust and effective solution for point cloud color reconstruction.
• The publicly available code facilitates further research and application in the field.