A Study On GPU-based Soft Shadows Technology

Shadows play an important role in real-time, photo-realistic rendering of computer-generated scenes. They can help to understand relative object position and size and provide useful visual cues that help in understanding the geometry of a complex receiver and occluder. Recently, with the fast development of programmable graphics hardware, GPU-Based real-time soft shadows rendering technology becomes the hot and difficult research points. This thesis is about real-time soft shadow technology using programmable graphics hardware and mainly focuses on two aspects: Firstly we implements and compares the newest real-time shadow generation algorithms and shadow anti-aliasing and soft shadow technology using shadow mapping. Secondly, this thesis presents an improved Percentage-Closer Soft Shadows algorithm, which is able to generate soft shadow for dynamic scenes in real-time and support dynamically changing light sources as well as viewpoints and different properties of materials in the scenes.This thesis introduces the shadow mapping mathematical principles and occluder distribution theory and elaborates how to change the shadow test function or approximation the occluder distribution to reduce aliasing and soften shadow edges. We research, implement and compare the shadow anti-aliasing technology and soft shadow rendering technology based on Percentage Closer Filter, Variance Shadow Maps, Convolution Shadow Maps and Exponential Shadow Maps. These algorithms only clamping the minimum filter width soften shadow edges and produce fixed-size shadow penumbra.The improved Percentage-Closer Soft Shadows algorithm has taken advantage of capability by choosing the filter width dynamically, based on an estimation of the "blocker depth". It can produce convincing shadow penumbrae the increase in size as the distance between the blocker and receiver increases. PCSS is based on the observation that as the size of the PCF kernel increases, the resulting shadows become softer. The challenge is to vary the filter size intelligently to achieve the correct degree of softness. The blocker search samples a search region by Poisson Disk Pattern, which can reduce banding without unduly increasing the number of samples. During filtering, variance shadow maps can be used to replace the standard shadow maps. It make it possible to pre-filter the shadow map by representing a probability distribution of depths at each shadow map texel. As to estimate the size of the penumbra using the computed blocker depth per pixel, summed-area tables allow us to vary the filter size dynamically, so summed-area variance shadow maps can be used directly, improving the quality and performance of the algorithm over the uniform grid percentage-closer filtering that was used in the original implementation. When rendering the scene, using a variable depth bias in the direction of the local depth gradient inside the filter kernel can eliminate surface acne and handle self-shadowing. So the improved PCSS technique can generate more perceptually accurate soft shadows.The improved Percentage-Closer Soft Shadows algorithm is suitable for a wide range of applications, from next-generation games to DCC/CAD applications. Because it uses just a single sample and required no special processing steps, the technique easily replaces existing shadow mapping shader code in an application while producing much more realistic images.