Adaptive and Feature-Preserving Bilateral Filters for Three-Dimensional Models

Bilateral filtering is a well-known tool to denoise or smooth one-dimensional (1D) signals, two-dimensional (2D) images, and three-dimensional (3D) models. The bilateral weights help preserve the edges or features more effectively than unilateral weights. However, it is immensely difficult to configure the scale parameters of the convolutional kernel functions. To overcome the difficulty, this paper proposes adaptive, feature-preserving bilateral filters for 3D models by introducing automatic, adaptive scale parameters. Firstly, the feature scale was defined on 3D models, bridging up the gap between feature scale and scale parameter of the Gaussian functions. Next, the scale descriptor was proposed to adaptively configure the scale parameter for each face of the target 3D model, changing the traditional approach of adopting the same scale parameter for all faces. On this basis, a feature-preserving local filter was designed by introducing the adaptive scale parameters to the iterative local scheme, and a modified global filter, which is robust to irregular sampling during denoising, was designed based on the adaptive scale parameters. The excellence of our filters was proved through experiments on multiple synthetic and real-world noisy models, in comparison to the state-of-the-art filters. The research results lay a solid basis for feature preservation and noise removal of 3D models.