Suavização de malhas usando filtros bilaterais adaptativos

Abstract

In 1998, Tomasi and Manduchi introduced the bilateral filter for grayscale and colo red images. It was subsequently extended to perform feature-preserving mesh denoising by Fleishman et al. and Jones et al. In this dissertation, we present a mesh smoothing algorithm using adaptive bilateral filters, which utilizes the common two-step architecture for mesh de noising. In the first step, we apply the bilateral filter to the mesh normal field, considering geodesic distances between centroids of adjacent faces. This strategy respects the spatial structure of the mesh itself by using intrinsic metrics. The parameters for the spatial and similarity weights of the algorithm are fully determined adaptively, avoiding the dilemma of previous methods that relied on trial and error to achieve satisfactory results. In the second step, the mean curvature flow, along with Sun et al.’s strategy, are combined to update the vertex positions so that the face normals correspond to the new normal field. That is, the direction of the vertex update is adjusted to a linear combination of the direction of the dif ferential coordinate and the direction given by the area-weighted average of the projections of the vertex onto the neighboring planes corresponding to the new filtered normals obtained in the first step, which improves the mesh quality. Extensive experiments demonstrate the effectiveness of the method in preserving mesh features while removing noise.