High Quality and Efficient Percentage Closer Soft Shadows

Percentage closer soft shadow (PCSS) is a prevalent algorithm in real-time applications due to its simplicity and efficiency.But soft shadows generated via PCSS often suffer from artifacts due to incorrect occluder estimation, meanwhile its performance drops down dramatically when the size of light source becomes large.Furthermore, the shadows also suffer from severe aliasing artifacts.In this paper we propose an efficient percentage closer soft shadow algorithm with smooth filtering. Multi-scale shadow maps are adopted to estimate search area accurately and calculate average blocker depth efficiently using a linear GPU quad-tree traversal algorithm.Finally, smooth soft shadows are generated via a high quality filtering algorithm.By integrating the proposed algorithm into a multiview shadow mapping framework, high quality shadows can be generated which approaches to the ground truth.The experimental results show that the proposed algorithm can generate smoother soft shadows with higher frame rates compared to the previous techniques.