Abstract: Curvature-based Volume rendering is a useful method to help scientists to find
features in data set and to design new mathematics model. In this paper, Fast GPU Raycasting
for curvature, in which data are reconstructed with tri-linear filtering and tri-cubic filtering and
derivatives, curvature and shading are computed on demand, is implemented in GPU. As a
result, computing on demand speeds up the method based on tri-cubic filtering; the method
based on tri-linear filtering is faster than the method based on tri-cubic filtering; the method
based on tri-cubic filtering can acquire more smooth result. We can select interactive camera
setting, transfer function and curvature function to highlight interesting feature with the method
based on tri-linear filtering and then render high quality image with the method based on
tri-cubic filtering.

Key words: computer application, GPU, raycasting, curvature