Isocube-based spherical wavelet for image-based relighting

摘要

Using the brute-force way of synthesising a scene that is, illuminated under a distant environment is computationally intensive. To speed up the synthesis, we can represent the appearance of the scene under different lighting directions as an illumination adjustable image (IAI). Relighting the scene under a lighting environment is then achieved by manipulating the pre-recorded images of the IAI. However, the relighting based on the pre-recorded images is still inefficient because of the large storage and computation requirements. The approaches based on spherical harmonics (SH) and cubemap-based spherical wavelets (CSW) have been proposed to improve the relighting efficiency. Note that the SH approach is suitable for lowfrequency lighting only, whereas the CSW approach suffers the problem of uneven sampling. In this paper, we propose a novel approach, Isocube-based spherical wavelets (IsoSW), to efficiently synthesise images based on the IAI representation. Unlike the SH approach, our approach can effectively capture high-frequency illumination effects. Compared with the CSW approach, our approach not only improves the rendering quality, but also makes the synthesis process more efficient.