Institute for Data Analysis and Visualization

This paper presents an efficient algorithm for filtering multi-layer depth images (MDIs) in order to produce approximate penumbras. The filtering is performed on a MDI that represents the view from the light source. The algorithm is based upon both ray tracing and the z-buffer shadow algorithm, and is closely related to convolution methods. The method's effectiveness is demonstrated on especially complex objects such as trees, whose soft shadows are expensive to compute by other methods. The method specifically addresses the problem of light-leaking that occurs when tracing rays through discrete representations, and the inability of convolution methods to produce accurate self-shadowing effects.