Stable and Fast Geometric Exaggeration via Saliency-Based Bilateral Filtering on GPU

Abstract

In this paper, we propose a GPU-based framework for reliably exaggerating the shape of a triangular mesh using mesh saliency and a boost filter. This framework is based on the high-boost mesh filter from digital signal processing, which affects the normal vectors of the triangular mesh and updates vertex positions to adapt to the exaggerated normal vectors. However, this process introduces noise at the vertices, and previous methods attempted to minimize this noise using an averaging filter. To address this issue, we apply a Bilateral filter to the high-boost filtering algorithm to effectively remove noise while exaggerating the mesh in the saliency direction, accelerating the process using GPU computation. Previous methods often resulted in noise, holes, or mesh shrinkage during the mesh enhancement process, whereas our method mitigates these issues. The effectiveness of our approach is evident not only in 3D objects but also in 3D-printed results. Through various experiments, we demonstrate that our method is an effective technique for exaggerating the shape of 3D meshes.