Supported by the Young Academic and Technical Leader Foundation of Yunnan Province of China under Grant No.2000YP20 (云南省中青年学术和技术带头人基金); Hong Kong Research Grant Council (RGC) Earmarked Research Grants under Grant No. CUHK4306/98E (香港研究资助局资助项目)
Illumination is an important aspect in realistic graphics rendering and most image related applications. The paper proposes a pure image based rendering method to synthesis the variation of illumination impacted on an object. Rather than estimating the BRDF function or any parameters in a reflectance model, Singular Value Decomposition (SVD) is used to estimate a low dimensional approximation of the image set of a Lambertian object under both changing of illumination and geometrical pose. In the method, light directions are analytically derived from the samples and the base images as well as images with known of the class of the objects. Images with any novel lighting directions can be rendered very efficiently by properly linear combination of the base images. Smooth morphing sequence of lighting and pose variation can be generated by linear interpolation of the SVD coefficients. A feature of this method is that it supplies a simple and compact way to represent dataset and synthesize novel images, which lends it to be suitable for many image-based applications.