基于线性G-buffer的后处理反走样

发布时间：2018-04-28 13:07

本文选题：延迟着色 + 反走样　；参考：《浙江大学》2016年硕士论文

【摘要】：图形是连续的。图形的光栅化过程,即用离散采样的像素表示连续的图形,必然会引起走样,具体表现为锯齿形边缘、动画中的闪烁、狭小物体的消失等现象。这些走样现象与图形真实感的要求产生了矛盾,因此图形反走样一直是实时绘制中的一个研究热点。本文首先概述了实时绘制中走样的理论基础,并介绍了基于硬件的各种反走样算法以及各种后处理反走样算法。后处理反走样算法主要包括基于图像分析的后处理反走样算法以及基于几何辅助的后处理反走样算法。基于图像分析的反走样算法的思想是通过图像处理方法,如启发式的形态学分析等方法,确定图像边界的长度和形状,然后进行图像滤波。该类方法简单高效,但是由于滤波方向受边界方向模板的约束,导致边界估计并不平滑。基于几何辅助分析的反走样算法的思想是通过存储场景的高分辨几何信息,准确地对场景的几何边界进行重建,但是这样会在G-buffer的存储和访问上增加开销,并且忽略了颜色纹理的走样。基于两类算法的互补性,本文提出了面向延迟着色的后处理统一反走样算法。我们绘制两次屏幕分辨率的场景几何信息,采用线性重构的采样生成G-buffer。通过判断深度法向差异将像素分为几何边界像素和非几何边界像素。对于几何边界像素,采用子像素重建反走样进行高质量的反走样滤波。对于非几何边界像素,提出了虚拟三角形的思想,利用虚拟三角形来为非几何边界像素的反走样处理提供几何信息,然后,使用改进的子像素级别的形态学反走样算法进行滤波。这样统一了尺度,避免因结合两类尺度不一致的算法而导致的交界处过渡不平滑。实验结果表明,本文算法有效地结合了基于几何辅助反走样处理的优势与基于图像分析反走样处理的优势,在基本保持反走样质量的基础上,降低了 G-buffer的存储开销,提高了反走样的效率。
[Abstract]:The graphics are continuous. The rasterization process of graphics, in which the discrete sampling pixels are used to represent the continuous graphics, will inevitably lead to the appearance of aliasing, such as sawtooth edges, flickering in animation, disappearance of narrow objects and so on. These aliasing phenomena are contradictory to the requirements of realistic graphics, so the anti-aliasing of graphics is always a hot topic in real time rendering. This paper first summarizes the theoretical basis of aliasing in real-time rendering, and introduces various anti-aliasing algorithms based on hardware and various post-processing anti-aliasing algorithms. Post-processing anti-aliasing algorithm mainly includes post-processing anti-aliasing algorithm based on image analysis and post-processing anti-aliasing algorithm based on geometric assistance. The idea of anti-aliasing algorithm based on image analysis is to determine the length and shape of image boundary by image processing methods such as heuristic morphological analysis and then filter the image. This method is simple and efficient, but the filtering direction is constrained by the boundary direction template, so the boundary estimation is not smooth. The idea of anti-aliasing algorithm based on geometric aided analysis is to reconstruct the geometric boundary of scene accurately by storing the high-resolution geometric information of the scene, but this will increase the cost of storage and access to G-buffer. And ignore the color texture aliasing. Based on the complementarity of the two algorithms, a post-processing unified anti-aliasing algorithm for delayed coloring is proposed. We draw the geometric information of the scene with twice screen resolution and generate G-buffer by linear reconstruction sampling. The pixels are divided into geometric boundary pixels and non-geometric boundary pixels by judging the depth normal difference. For geometric boundary pixels, high quality anti-aliasing filtering is performed by subpixel reconstruction. For non-geometric boundary pixels, the idea of virtual triangles is proposed. Virtual triangles are used to provide geometric information for anti-aliasing processing of non-geometric boundary pixels. The improved subpixel level morphological anti-aliasing algorithm is used to filter. In this way, the scale is unified and the boundary transition is not smooth due to the combination of the two kinds of scale inconsistent algorithms. Experimental results show that the proposed algorithm effectively combines the advantages of geometric aided anti-aliasing processing and image analysis-based anti-aliasing processing, and reduces the storage overhead of G-buffer on the basis of maintaining the anti-aliasing quality. The efficiency of anti-aliasing is improved.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】