没有合适的资源?快使用搜索试试~ 我知道了~
温馨提示
Distribution ray tracing uses Monte Carlo integration to solve the rendering equation. This technique was introduced by Cook et. al, and was notable because of its simplicity and its ability to simulate areal luminaires, camera lens e ects, motion blur, and imperfect specular re ection[5]. Distribution ray tracing has been extended and modi ed by many researchers, most no- tably by Kajiya who added true indirect illumination[12]. Distribution ray tracing has also been used as the viewing component of radiosity systems (e.g [37, 3]). In this paper we examine some of the central issues of distribu- tion ray tracing that have been overlooked in the literature but are still of importance
资源推荐
资源详情
资源评论
Distribution Ray Tracing: Theory and
Practice
Peter Shirley
Changyaw Wang
Computer Science Department
Indiana University
Proceedings of the Third Eurographics Workshop on Rendering
1 Intro duction
Distribution ray tracing uses Monte Carlo integration to solve the rendering
equation. This technique was introduced by Cook et. al, and was notable
because of its simplicity and its ability to simulate areal luminaires, camera
lens eects, motion blur, and imp erfect specular reection[5]. Distribution
ray tracing has been extended and modied by many researchers, most no-
tably by Ka jiya who added true indirect illumination[12]. Distribution ray
tracing has also b een used as the viewing comp onent of radiosity systems
(e.g [37, 3]). In this paper we examine some of the central issues of distribu-
tion ray tracing that have been overlooked in the literature but are still of
importance.
Our interest in distribution ray tracing arises for three basic reasons.
The rst reason is that we b elieve any rendering system that has imp erfect
specular reection (e.g. brushed steel) will use distribution ray tracing as a
viewing method. Second, distribution ray tracing is a natural way to render
outdoor scenes, where the eects of single and double reection dominate
color, and where the geometry is extremely complex. Third, it is p ossible to
run ray tracing on large distributed memory multiprocessors[25].
Although the implementation of a simple distribution ray tracer is usually
implied to be straightforward, there are several important implementational
1
issues that have not been discussed in the literature, and the relationship be-
tween distribution ray tracing code and distribution ray tracing theory has
never b een completely stated. In this pap er, we present some implementa-
tional and mathematical details that we have found to be important in the
design of our software. In Section 2, we discuss the basics of Monte Carlo In-
tegration and Quasi-Monte Carlo integration. In Section 3, we apply Monte
Carlo integration to pixel ltering and use this as an example of how the
mathematics inuence the design of distribution ray tracing co de. Section 4,
we discuss the illumination calculations in a distribution ray tracer. This sec-
tion argues that illumination calculation is not well understo od and discusses
some problems that need further research. Section 5 discusses the need for
perception based display of the output of a distribution ray tracer. Section 6
discusses several implementational considerations we feel are non-obvious and
have recently forced us to redesign our software. Finally, Section 7 summa-
rizes our results and current plan of research.
2 Monte Carlo Integration and Quasi-Monte
Carlo Integration
In Monte Carlo integration, random points with some distribution are used
to nd an approximate value for the integral. For an integral
I
of a function
f
over some space
S
, we can generate an estimate of
I
using some set of
random points
1
through
N
, where each
i
is distributed according to a
probability density function
p
:
I
=
Z
x
2
S
f
(
x
)
d
x
N
X
i
=1
f
(
i
)
p
(
i
)
(1)
Although distribution ray tracing is usually phrased as an application of
Equation 1, many researchers replace the
i
with more evenly distributed
(quasi-random) samples (e.g. [4, 19 ]). This approach can b e shown to be
sound by analyzing decreasing error in terms of some discrepancy measure[44,
43, 19, 28] rather than in terms of variance. However, it is often convenient
to develop a sampling strategy using variance analysis on random samples,
and to instead use non-random, but equidistributed samples in an implemen-
tation. This approach is almost certainly correct, but its justication and
implications have yet to be explained.
2
We often have integrals that take the form of a strictly p ositive weighted
average of a function:
I
=
Z
x
2
S
w
(
x
)
f
(
x
)
d
x
where
w
is a weighting function with unit volume. To solve this by Equa-
tion 1, the optimal choice for the probability function is
p
(
x
) =
C w
(
x
)
f
(
x
),
but as is often pointed out, this choice requires us to already know the value of
I
. Instead, people often either choose uniform
p
, or set
p
(
x
) =
w
(
x
)[4, 24 , 16 ].
In graphics we usually rep eatedly perform an integral using many dierent
f
chosen from some set of functions
F
. To decide which
p
to use, we could
try to minimize the average variance of our estimate across all
f
2
F
. We
cannot do this with such a vague denition of
F
, but we can approximate
our situation in graphics by assuming that we are no more likely to have
large values of
f
in a particular p oint in
S
than any other. This leads to the
conclusion that the average value of
f
2
(
x
) at any given
x
is some constant
< f
2
>
. The average variance of the estimator
I
0
=
w
(
)
f
(
)
=p
(
) is then
just:
v ar
ave
(
I
0
) =
Z
x
2
S
w
2
(
x
)
< f
2
>
p
(
x
)
d
x
?
I
2
Which can b e shown by calculus of variations to be minimal when
p
=
w
. This implies that the intuitively appealing choice of
p
=
w
has some
theoretical justication as well.
3 Pixel Filtering
The color of a pixel
I
(
i; j
) can b e expressed as an integral:
I
(
i; j
) =
Z
p
2
S
w
(
p
)
L
(
p
)
dA
p
(2)
where
p
is a point on the viewp ort (or lmplane if a camera model is used),
L
(
p
) is the radiance seen through the viewp ort at
p
, and
S
is the non-zero
region of the lter function
w
.
Like other parts of the full rendering equation, Equation 2 can be solved
by Monte Carlo Integration. Rewriting with the assumption that the same
origin-centered weighting function is used for every pixel yields the estimator
3
剩余17页未读,继续阅读
资源评论
qq_26493017
- 粉丝: 0
- 资源: 12
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- VCSEL芯片,中国前5强生产商排名及市场份额(by QYResearch).pdf
- 办公型钢笔,全球前11强生产商排名及市场份额(by QYResearch).pdf
- 车库卷帘门,全球前16强生产商排名及市场份额(by QYResearch).pdf
- 机器人减速机,全球前10强生产商排名及市场份额(by QYResearch).pdf
- 低温真空泵,全球前8强生产商排名及市场份额(by QYResearch).pdf
- 钢制气瓶,全球前5强生产商排名及市场份额(by QYResearch).pdf
- 加工油,全球前10强生产商排名及市场份额(by QYResearch).pdf
- 瓶装水灌装设备,全球前7强生产商排名及市场份额(by QYResearch).pdf
- 氢燃料电池无人机,全球前8强生产商排名及市场份额(by QYResearch).pdf
- 全球功能纺织面料产业链,现状和趋势.pdf
- 全球轨道交通信号检测和监测各地区市场现状分析.pdf
- 全球吹雪机产业链,现状和趋势.pdf
- 全球轨道交通检测和监测市场现状分析.pdf
- 输送机托辊,全球前20强生产商排名及市场份额.pdf
- 全球轨道交通通信检测和监测各地区市场现状分析.pdf
- 压电堆叠执行器,全球前14强生产商排名及市场份额(by QYResearch).pdf
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功