A Joint Anti-Aliasing Method for Gaussian Splatting Based on Wavelet-Guided Supersampling and Depth-Adaptive Weighted Pre-Filtering

Due to insufficient sampling, 3D Gaussian Splatting (3DGS) is prone to aliasing artifacts during rendering, especially under significant viewpoint changes or in high-frequency regions, often resulting in detail loss, jagged edges, and visual artifacts. To address this issue, this paper proposes a joint anti-aliasing method based on wavelet-guided super-sampling and depth-adaptive weighted pre-filtering. Specifically, a multi-scale wavelet decomposition is applied to the input image in a preprocessing stage to extract high-frequency information, which reveals regions with rapid intensity changes such as edges or fine textures. These high-frequency regions are then mapped to corresponding Gaussian ellipsoids in the 3D representation. During the optimization process, the identified high-frequency Gaussians are subdivided to enable supersampling around detailed structures, thereby refining high-frequency geometry and mitigating aliasing artifacts. In the rendering stage, a distance-based weighting mechanism is introduced, which serves as a form of dynamic pre-filtering across different depths. This allows Gaussians closer to the camera to contribute more significantly to the final pixel color, enhancing the realism and depth perception of the rendered image. Experimental results demonstrate that the proposed method effectively reduces aliasing and improves detail preservation and overall rendering quality across various complex scenes.