Post-Processing Anti-Aliasing Based On Linear G-buffer

Graphics is continuous.Graphics rasterization will cause aliasing phenomena,ie.,representing continuous graphics as discrete pixels,such as jagged edges,blinking in animation and so on.Therefore,anti-aliasing is always one of fundamental problems in computer graphics.In this thesis,we first briefly review the basic theory and fundamental concepts about aliasing in real-time rendering.Then we introduce the hardware-based anti-aliasing algorithms and post-processing anti-aliasing algorithms for deferred shading.The post-processing anti-aliasing algorithms are classified into two categories:the image-based approach and the geometry-based approach.The image-based approach detects aliasing features in a screen-resolution image and then performs heuristic filtering.The approach exhibits good performance,but because the geometric information is insufficient,the approach tends to over-blur some visual features.The geometry-based approach stores the high resolution geometry information,ie.,G-buffer.Then it reconstructs the sub-pixel information according to the geometry similarity among the sub-pixels and its neighboring shading samples,and performs filtering to alleviate geometry aliasing.However,the high resolution G-buffer will increase memory storage and access overheads.Based on advantages of two approaches above,we present a unified anti-aliasing algorithm to deal with both geometry aliasing and shading aliasing.We linearly reconstruct a G-buffer via two screen resolution G-buffers,and classify all pixels into geometric boundary pixels and non-geometric boundary pixels according to the difference between depth/normal For the former pixels,it can achieve high-quality geometric feature anti-aliasing via sub-pixel reconstruction.Forthe latter pixels,we adopt the virtual triangle as sub-pixel-level geometric information,and improve the morphological method as a sub-pixel version to deal with the aliasing.Because of the virtual triangle adopted,we could avoid the unsmooth transition between screen pixels with different algorithms.The results show that the algorithm effectively combines the advantages of image-based approach and geometry-based approach.Without sacrifice of anti-aliasing quality,the proposed approach achieves higher performance with smaller size of G-buffer than previous method.