CN102496171A - Hierarchical replacement method for (t, s) sequences - Google Patents
Hierarchical replacement method for (t, s) sequences Download PDFInfo
- Publication number
- CN102496171A CN102496171A CN 201110400068 CN201110400068A CN102496171A CN 102496171 A CN102496171 A CN 102496171A CN 201110400068 CN201110400068 CN 201110400068 CN 201110400068 A CN201110400068 A CN 201110400068A CN 102496171 A CN102496171 A CN 102496171A
- Authority
- CN
- China
- Prior art keywords
- sequence
- integral
- mod
- niederreiter
- further comprise
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
Abstract
The invention provides a computer image rendering method. According to the method, the position of sampling points is determined by hierarchical replacement operation based on (t, s) sequences. Compared with a Latin hypercube sampling method, the computer image rendering method has the advantages that: the consumption of internal memory during running of a program can be remarkably reduced; the quantity of the sampling points is not required to be predetermined; and the application range is remarkably expanded.
Description
Technical field
The present invention relates to a kind of computer picture rendering intent, particularly a kind of rendering intent that utilizes quasi-random numbers (quasi-random numbers) to confirm sampling point position.
Background technology
Play up the field at computer picture, for simulating various effect of shadow, for example, the depth of field, motion blur, global illumination etc., need carry out evaluation to the integration with following form:
Wherein I is the unit interval, function f () by the contextual data decision of the effect that will simulate and input.Formula (1) generally can't accurately be found the solution, and in the render process of reality, always is to use the method for numerical evaluation to ask its approximate value:
The method of confirming sampling point position
in the formula (2) can be divided into two types; One type of method that is to use random number, the another kind of method that is to use quasi-random numbers.Quasi-random numbers is a kind of distribution pattern with satisfactory texture, and the efficient higher than random number can be provided in the Integral Estimation of formula (2).
(t, s) sequence is a kind of main quasi-random numbers type, generation b system (t, s) method of the j of i some dimension component is following in the sequence:
Wherein
For (t, s) j of sequence supports one's family into matrix (generator matrices), its coefficient
Be Galois field F
bMiddle element.b
r(i) the r bit digital of expression integer i under the b system is represented, promptly
M is selected maximal accuracy under the b system is represented.
(t, s) example of sequence comprises: Sobol sequence, Faure sequence, Niederreiter sequence and Niederreiter-Xing sequence etc.It is different that (t, s) difference of sequence is that the building method of its generator matrix is different.
Although quasi-random numbers can provide the efficient higher than random number, along with the increase of dimension, the quality of quasi-random numbers can descend gradually, and random number can not.For fully utilizing quasi-random numbers and random number advantage separately; Art B.Owen; " Latin Supercube Sampling for Very High-Dimensional Simulations ", ACM Transactions on Modeling and Computer Simulation, vol.8; No.1; Jan.1998, pp.71-102. provide the method for the super cube sampling of a kind of Latin (Latin Supercube Sampling), the method is characterized in that through quasi-random numbers being carried out the higher-dimension component that random permutation generates sampled point.
Because what adopt is common replacement operator, realizes that the computer program of the super cube sampling of Latin need consume and the proportional memory size of the quantity of sampled point when operation, thereby be unwell to the task of playing up of a large amount of sampled points of needs.On the other hand, this common replacement operator needs to confirm in advance the quantity of sampled point, thereby can't be used for asymptotic expression and play up (progressive rendering).
Summary of the invention
The invention provides a kind of computer picture rendering intent, this method is based on (t, s) sequence are confirmed the position of sampled point through the hierarchical replacement operator.
Realize the scheme of technical purpose of the present invention, may further comprise the steps:
The integration that needs evaluation during (A) for rendering image
Use the method for numerical evaluation to ask its approximate value
(B) confirm the method for the value of variable described in the step (A)
; 0≤i≤N-1 wherein may further comprise the steps:
Will
Be grouped into
P wherein
0=0, p
Q+1=s, 0<p
V+1-p
v≤s
Max, 0≤v≤q, s
MaxBe selected largest;
H wherein
j(i) :=(b
-1, b
-2..., b
-M) C
(j)(b
0(i), b
1(i) ..., b
M-1(i))
T,
Be that (t, s) j of sequence supports one's family into matrix to a b system, b
r(i) the r bit digital of expression integer i under the b system is represented, promptly
M is selected maximal accuracy under the b system is represented;
For set of variables
1≤v≤q wherein, confirm that the method for its value may further comprise the steps:
i(v)=π
v(i),1≤v≤q;
Wherein
ξ wherein
v, ξ
V, 0, ξ
V, 1..., ξ
V, b-1, ξ
V, 0,0..., ξ
V, b-1, b-1..., ξ
V, b-1..., b-1,1≤v≤q is separate random number.
The present invention has the following advantages:
(t, s) sequence are confirmed the position of sampled point to compare the memory consumption when the present invention can significantly reduce program run with the super cube sampling of Latin through the hierarchical replacement operator owing to the present invention is based on.
(t, s) sequence are confirmed the position of sampled point through the hierarchical replacement operator, need not confirm the quantity of sampled point in advance, sample with the super cube of Latin and compare, and the scope of application of the present invention significantly strengthens owing to the present invention is based on.
Embodiment
Below in conjunction with embodiment the present invention is elaborated.
Table 1 has been listed the C++ example code that calculates 2 system Niederreiter sequences:
Table 1
In code shown in the table 1, maximal accuracy is made as 32.The reason of setting a maximal accuracy is, in numerical evaluation, and both can not the also unnecessary calculating of carrying out infinite precision.
In code shown in the table 1, largest is made as 12 dimensions.The effect of setting largest is, on the one hand, calculate (t, s) data of the generator matrix that need use of sequence need be stored in the computer-readable medium in advance, and any equipment in the reality all can not the unlimited many data of storage; On the other hand, though the quality of quasi-random numbers is superior to random number, along with the increase of dimension, the quality of quasi-random numbers can descend gradually, thus in reality the infinite dimensional quasi-random numbers of also unnecessary use.
(choosing of largest will decide according to the maximal accuracy of its mass parameter (qualty parameters) t and setting for t, s) sequence, and its principle is that the mass parameter of selected largest should not surpass the maximal accuracy of setting to different.
Table 2 has been listed 2 system Sobol sequences, Niederreiter sequence and the mass parameter of Niederreiter-Xing sequence in different dimensions.The little expression quality of this more parameters is high more, and minimum is 0.
Table 2
Dimension | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 20 |
Sobol | 0 | 0 | 1 | 3 | 5 | 8 | 11 | 15 | 19 | 23 | 27 | 31 | 35 | 40 | 45 | 71 |
Niederreiter | 0 | 0 | 1 | 3 | 5 | 8 | 11 | 14 | 18 | 22 | 26 | 30 | 34 | 38 | 43 | 68 |
Niederreiter-Xing | 0 | 0 | 1 | 1 | 2 | 3 | 4 | 5 | 6 | 8 | 9 | 10 | 11 | 13 | 15 | 21 |
Can find out from last table; For 2 system Sobol sequences and Niederreiter sequence, if maximal accuracy is made as 32, then largest should be elected 12 dimensions as; Because the Sobol sequence of 12 dimensions and the mass parameter of Niederreiter sequence are respectively 31 and 30; Do not surpass 32, and the Sobol sequence of 13 dimensions and the mass parameter of Niederreiter sequence are respectively 35 and 34, have surpassed 32.
Use (t, s) sequence need be set a largest, when the dimension of the required sampled point of rendering image surpasses this setting value; Need be to (t; S) sequence number of sequence is carried out a hierarchical replacement operator, and the sequence number after the displacement can be used for, and former (t, s) sequence generates new component.Table 3 has been listed the C++ example code of realizing this hierarchical replacement operator:
Table 3
Can find out from code shown in the table 3; Different with the common replacement operator that the super cube sampling of Latin is adopted; Realize that its memory consumption of computer program of hierarchical replacement operator and the quantity of sampled point have nothing to do; Compare with the super cube sampling of Latin, the memory consumption during program run significantly reduces.
Table 4 has been listed based on 2 system Niederreiter sequences, confirms the C++ example code of sampling point position through the hierarchical replacement operator, has wherein used at code shown in table 1 and the table 3:
Table 4
Table 5 has been listed the C++ example code of code shown in the use table 4 in render process:
Table 5
Can find out from code shown in table 4 and the table 5; Different with the common replacement operator that the super cube sampling of Latin is adopted; The hierarchical replacement operator need not be confirmed the quantity of sampled point in advance; The position that only need provide corresponding sequence number can generate required sampled point, this point for gradual play up particularly important.In gradual playing up, renderer feeds back current rendering result in real time and gives the user, when stops playing up task by user's decision, thereby can't confirm the quantity of sampled point in advance, can only rely on the sequence number of sampled point to generate the position of sampled point.The hierarchical replacement operator can satisfy gradual this requirement of playing up fully.
In code shown in the table 4, to each divide into groups all to have used identical (t, s) sequence of all dimensions of sampled point component; I.e. 2 system Niederreiter sequences; Remove to use identical (t, s) sequence also can be used radix identical but (t, s) sequence that type is different outward; For example, 2 system Niederreiter-Xing sequences.Though the quality of Niederreiter-Xing sequence is superior to Sobol sequence and Niederreiter sequence; But there is a restriction in the use; Promptly must confirm total sampling dimension in advance; Thereby suitable play up as image in a kind of blanket method, if but the dimension of current group is known and do not have the more sampling of higher-dimension, then available Niederreiter-Xing sequence replacement Niederreiter sequence is to obtain better rendering effect.
It is a kind of that (t, s) expansion of sequence is referred to as randomization (t, s) (randomized (t, s) sequences) its role is to estimate the error size of integral approach value to sequence.Though needn't go to estimate the error size of integral approach value during rendering image; But this method of randomization is used in the flaw of alleviating in the playing up of animation in the image; The pictures different frame uses different random numbers that sampled point is carried out randomization in promptly animation being played up, to desalinate the flaw pattern that possibly exist.
Table 6 has been listed the C++ example code that on the basis of code shown in the table 4, has increased the randomization step:
Table 6
Claims (4)
1. a computer picture rendering intent is characterized by, and may further comprise the steps:
The integration that needs evaluation during (A) for rendering image
Use the method for numerical evaluation to ask its approximate value
(B) confirm the method for the value of variable described in the step (A)
; 0≤i≤N-1 wherein may further comprise the steps:
Will
Be grouped into
P wherein
0=0, p
Q+1=s, 0<p
V+1-p
v≤smax, 0≤v≤q, smax are selected largest;
H wherein
j(i) :=(b
-1, b
-2..., b
-M) C
(j)(b
0(i), b
1(i) ..., b
M-1(i))
T,
Be that (t, s) j of sequence supports one's family into matrix to a b system, b
r(i) the r bit digital of expression integer i under the b system is represented, promptly
M is selected maximal accuracy under the b system is represented;
For set of variables
1≤v≤q wherein, confirm that the method for its value may further comprise the steps:
i(v)=πv(i),1≤v≤q;
ξ wherein
v, ξ
V, 0, ξ
V, 1..., ξ
V, b-1, ξ
V, 0,0..., ξ
V, b-1, b-1..., ξ
V, b-1 ..., b-1, 1≤v≤q is separate random number.
2. like the said method of claim 1., also comprise, each of all dimensions of sampled point component divided into groups to use identical (t, s) identical but (t, s) sequence that type is different of sequence or radix.
3. like the said method of claim 1., wherein, said (t, s) sequence comprises Sobol sequence, Faure sequence, Niederreiter sequence and Niederreiter-Xing sequence.
4. like the said method of claim 1., wherein, said (t, s) sequence also comprises randomized (t, s) sequence.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110400068 CN102496171A (en) | 2011-12-06 | 2011-12-06 | Hierarchical replacement method for (t, s) sequences |
PCT/CN2012/084742 WO2013082993A1 (en) | 2011-12-06 | 2012-11-16 | Hierarchical replacement method for (t, s) sequence |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN 201110400068 CN102496171A (en) | 2011-12-06 | 2011-12-06 | Hierarchical replacement method for (t, s) sequences |
Publications (1)
Publication Number | Publication Date |
---|---|
CN102496171A true CN102496171A (en) | 2012-06-13 |
Family
ID=46187993
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN 201110400068 Pending CN102496171A (en) | 2011-12-06 | 2011-12-06 | Hierarchical replacement method for (t, s) sequences |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN102496171A (en) |
WO (1) | WO2013082993A1 (en) |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101763636B (en) * | 2009-09-23 | 2012-07-04 | 中国科学院自动化研究所 | Method for tracing position and pose of 3D human face in video sequence |
CN101923590B (en) * | 2010-08-16 | 2012-10-03 | 北京理工大学 | High-efficiency Latin hypercube experimental design method |
-
2011
- 2011-12-06 CN CN 201110400068 patent/CN102496171A/en active Pending
-
2012
- 2012-11-16 WO PCT/CN2012/084742 patent/WO2013082993A1/en active Application Filing
Also Published As
Publication number | Publication date |
---|---|
WO2013082993A1 (en) | 2013-06-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Gurarslan et al. | Numerical solution of advection-diffusion equation using a sixth-order compact finite difference method | |
JP2011041326A5 (en) | ||
CN103746774A (en) | Error resilient coding method for high-efficiency data reading | |
Picasso et al. | A numerical study of some Hessian recovery techniques on isotropic and anisotropic meshes | |
CN105183880A (en) | Hash join method and device | |
CN101464995B (en) | Digital watermarking method for two-dimension vector space data based on spacing topological | |
CN101127911A (en) | A block size computing method and transmission method for cylinder panorama video code | |
Wang et al. | A new perturbation method to the Tent map and its application | |
CN102496170A (en) | Method for decomposing sampling task | |
CN110188407A (en) | The determination method and device of fluid flow parameter in porous media | |
CN111160560A (en) | Method and system for predicting resources required by analog quantum computation | |
CN102496172A (en) | Method for generating high-dimensional weight of sampling point | |
CN102496171A (en) | Hierarchical replacement method for (t, s) sequences | |
US8255194B2 (en) | Judiciously retreated finite element method for solving lubrication equation | |
CN102496173A (en) | Digimigration method used for separating sampling points | |
Fernández et al. | Algorithm for normal random numbers | |
CN105426488A (en) | WebGL based large-batch oblique photography measurement three-dimensional model loading method | |
Crittenden | Igloo pixelizations of the sky | |
CN103646376A (en) | Digital watermark image generation method | |
Tang et al. | A non‐oscillatory multimoment finite‐volume global transport model on a cubed‐sphere grid using the WENO slope limiter | |
CN106650126B (en) | Acceleration method for calculating array layout resistance network | |
JP2007041199A5 (en) | ||
Cornel et al. | Integrated Simulation and Visualization for Flood Management | |
Fulman | An inductive proof of the Berry-Esseen theorem for character ratios | |
Du et al. | The Aztec Diamond edge-probability generating function |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20120613 |