CN106898038B - A kind of rendering method merging HM filter using light frequency curve - Google Patents
A kind of rendering method merging HM filter using light frequency curve Download PDFInfo
- Publication number
- CN106898038B CN106898038B CN201710054403.8A CN201710054403A CN106898038B CN 106898038 B CN106898038 B CN 106898038B CN 201710054403 A CN201710054403 A CN 201710054403A CN 106898038 B CN106898038 B CN 106898038B
- Authority
- CN
- China
- Prior art keywords
- pixel
- filter
- rendering
- image
- light
- 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.)
- Expired - Fee Related
Links
- 238000009877 rendering Methods 0.000 title claims abstract description 68
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000001914 filtration Methods 0.000 claims description 13
- 238000005070 sampling Methods 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 3
- 239000000523 sample Substances 0.000 description 58
- 238000009826 distribution Methods 0.000 description 21
- 230000003044 adaptive effect Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000005286 illumination Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000012804 iterative process Methods 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000037361 pathway Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000005295 random walk Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
- G06T15/06—Ray-tracing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Graphics (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of iterative Adapti ve rendering methods for merging HM filter using light frequency curve, comprising: step 1, is tentatively rendered using less light sample size to image, carries out noise remove to rendering image using HM filter；Step 2, the mean square error of current HM filter is estimated using SURE, and calculates the best light sample size of each pixel in rendering image；Step 3, according to the amount of noise in rendering image, the threshold value of HM filter is updated；Step 4, rendering is re-started to image according to best light sample size, and noise remove is carried out using updated HM filter threshold value；Step 5, step 2~step 4 is repeated, until rendering image is met the requirements.Rendering method provided by the invention can save a large amount of computing cost, guarantee to reach time saving purpose while rendering quality.
Description
Technical field
The present invention relates to computer visual image Rendering fields, and in particular to a kind of to be merged using light frequency curve
The iterative Adapti ve rendering method of HM filter.
Background technique
There is very long phase of history using Monte Carlo ray tracing technology reconstructed image.Current work is in this neck
Domain also achieves bigger progress, but main target is still identical: that is, going to mention using small number of light sample
High picture quality shortens to improve efficiency and generates the time.However the image generated using very few light sample is easy to be made an uproar
Sound shadow is rung, it is desirable that has corresponding efficient image filter to obtain smooth image to remove noise.
Monte Carlo rendering is used to generate the time required for high quality graphic in order to shorten, general there are two types of basic plans
Slightly, i.e. Adapti ve rendering and rendering post-processes.The core concept of Adapti ve rendering is in render process according to specific image district
The complexity in domain adjusts the light sample size at each pixel.Rendering post-processing is then after rendering is completed, to rendering
Obtained result images are handled, and primary operational includes to filter or to sample, picture element interpolation.
Adapti ve rendering refers mainly to two aspects, i.e., adaptively sampled and adaptive filtering.Adaptively sampled is in error
Biggish image-region increases sample rate, that is, improves the quantity of sampling light in the biggish region of pixel value variance.It is adaptive
The average value that its neighborhood is taken to specified pixel should be filtered, to remove noise spot, the pixel value in this smooth block region.
Adaptive filtering is smaller than adaptively sampled required computing cost in practical applications, but bright to edge
The method picture quality obtained of the image adaptive filtering aobvious, characteristic is strong is simultaneously bad.
Summary of the invention
The present invention provides a kind of iterative Adapti ve rendering method for merging HM filter using light frequency curve, energy
Enough accelerate Monte Carlo rendering, while obtaining good picture quality.
A kind of iterative Adapti ve rendering method merging HM filter using light frequency curve, comprising:
Step 1, image is tentatively rendered using less light sample size, using HM filter to rendering image into
Row noise remove；
Step 2, the mean square error of current HM filter is estimated using SURE, and calculates each pixel in rendering image
Best light sample size；
Step 3, according to the amount of noise in rendering image, the threshold value of HM filter is updated；
Step 4, rendering is re-started to image according to best light sample size, and uses updated HM filter threshold
Value carries out noise remove；
Step 5, step 2~step 4 is repeated, until rendering image is met the requirements.
In step 1, for sample light sample size according to being set, less is a relative concept, can be directed to institute
The image rendering quality needed is set, appropriate to increase light sample size when image rendering quality requirement is higher.
The HM filter measures the similarity degree of pixel using the frequency curve of light sample at pixel, i.e., non local
A kind of HM (Histogram for merging pixel using light sample frequency distribution curve is proposed on the basis of mean filter
Merge) filter, and noise removal process is carried out to the image that Monte Carlo renders using the filter, to improve wash with watercolours
Contaminate the quality of image.
The error that the present invention generates HM filter using SURE (Stein ' s Unbiased Risk Estimator) into
Row estimation, obtains the best light sample size at each pixel, is then carried out again using best light sample size to image
Secondary rendering, HM filter adjusts threshold value according to the amount of noise in rendering image, carries out noise remove, by iterative adaptive
Rendering method is answered, a large amount of computing costs can be saved, while guaranteeing to render quality, reaches time saving purpose.
Preferably, carrying out noise remove to rendering image using following formula as unit of dough sheet:
In formula:For the color value for rendering pixel x in image after removal noise；
W is half length of dough sheet put centered on pixel x；
VyFor the result after the dough sheet denoising centered on pixel y；
Y-x is distance of the pixel y away from pixel x.
Preferably, calculating best light sample size a using following formula:
In formula: m is the budget of light sample size；
S (x) is the light sampling function at pixel x；
S (y) is the light sampling function at pixel y；
For the set for meeting the following conditions pixel y:
Dough sheet P centered on pixel yyMeet
In formula:For the dough sheet P centered on pixel xxWith the dough sheet P centered on pixel yyDistance；
K is the threshold value of HM filter.
Preferably, the calculation formula of light sampling function S (x) is as follows:
In formula: VxFor the result after the dough sheet denoising centered on pixel x；
F(Vx) it is the filters filter function for filtering the dough sheet centered on pixel x；
SURE(F(Vx)) for using the mean square error of the SURE current HM filter calculated；
E2It (x) is the evaluated error of the color value of pixel x；
I(F(Vx))2It is the color value of pixel x after filtering；
∈, which is one, prevents the minimum numerical constant that divisor is zero.
Preferably, using PSNR measure rendering picture quality, set iteration ends threshold value asCurrent rendering image phase
The PSNR variable quantity of rendering image is less than when completing than last iterationWhen, iteration ends.
The iterative Adapti ve rendering method provided by the invention for merging HM filter using light frequency curve, Ke Yijie
A large amount of computing cost is saved, is guaranteeing to reach time saving purpose while rendering quality.
Detailed description of the invention
Fig. 1 is the correspondence diagram of light sample of color and pixel samples distribution of color；
Fig. 2 is the light sample of color point that three different pixels in image are generated using Monte Carlo path tracing algorithm
Cloth.
Specific embodiment
With reference to the accompanying drawing, merge the iterative Adapti ve rendering side of HM filter using light frequency curve to the present invention
Method is described in detail.
As shown in Figure 1, a kind of iterative Adapti ve rendering method for merging HM filter using light frequency curve, including
Following steps:
Step 1, image is tentatively rendered using less light sample size, using HM filter to rendering image into
Row noise remove.
Light propagation can be expressed in the form of space path, i.e., propagated just by calculating the light of each independent pathway
It can estimate global illumination.According to following space path integral formula, the color of each pixel is by all possible opticpath
Integral obtains:
Herein: ΩxIndicate all opticpaths at pixel x；P is the path of random length；Function f (p) indicates illumination
Path is the illumination contributions of p；D μ (p) is the calculation of opticpath.
According to above-mentioned formula, the color of image at pixel x can be by nxA random road obtained using Monte Carlo
DiameterEstimation obtains.IfIt indicates by random walkThe color of propagation is (i.e.), then cover spy
Carlow method can be represented as the approximate solution of u (x):
So, Monte Carlo approximate error n (x) can be represented as:
Render time can be controlled to feasible method within an acceptable range again are as follows: reduce light while reducing approximate error
Line number of samples, then using rendering post-processing (i.e. filtering image).Filtering can be substantially reduced variance, but can also be substantially improved
Approximate deviation, the filter type that unique one kind will not introduce deviation is exactly pixel of the combination with same characteristic features, that is, is possessed
The pixel x of same pixel color u (x).But two pixel x, pixel y are determined based on unknown pixel color u (x) and u (y)
Whether color is identical is nearly impossible, and feasible way is the light sample of color of both expectationsFollow similar distribution.
In addition, if N number of pixel possesses the distribution of identical light sample of color, then the set of these samples can be by
Regard one N times big of the superset for following certain potential distribution as, as long as being simply averaged to this N number of pixel, their side
Difference will reduce N times.
The color of pixel and the distribution relation of light sample of color as shown in Figure 1, every light sample in a certain pixel
It can show corresponding light sample of color, all light sample of color superpositions, the as color of the pixel on a certain pixel.
The sample of color experience distribution map at given pixel is considered first, as shown in Fig. 2, chasing after using Monte Carlo path
Track algorithm generates the light sample of color distribution of three different pixels in image.The color for three pixels chosen in figure connects very much
Closely, the sample (two frames of lower section) of two of them pixel roughlys abide by same light sample of color distribution, and this light
The distribution of line sample of color is very big with the light sample of color distributional difference of one other pixel (frame of the top).This shows even if picture
Plain color is quite similar, and the additional information that the distribution of light sample of color provides can also help to distinguish the different picture of other features
Element.
In the present invention, useThe light sample of color collection of pixel x is indicated, with h (x) come table
Show corresponding experience distribution of color (i.e. pixel color distribution).In order to measure pixel similarity, use the distribution of cabinet experience as
The description of pixel color distribution.Since what is usually handled is image three-colo(u)r, a single three-dimensional color space both can be used
Three one-dimensional color frequency curves (one channel of each color) also can be used in frequency curve.
Assuming that the light sample of color at pixel x is Cx, the light sample of color at pixel y is Cy, the corresponding n of the twobCase
Body distribution is expressed as nbDimensional vector respectively indicates as follows:
Based on Chi-Square distance, it then follows standard below:
Herein: nx=∑ihi(x), ny=∑ihi(y), corresponding light total sample number amount is respectively indicated, k (x, y) is h
(x) in+h (y) non-empty cabinet quantity.
In view of spatial continuity, above-mentioned pixel distance be can be extended as centered on x and y, the dough sheet of half a length of w,
It is expressed as follows:
In formula: PxIt represents centered on pixel x, is half long dough sheet with w；
PyIt represents centered on pixel y, is half long dough sheet with w；
T indicates the point centered on pixel x, is half long with w, the pixel on the dough sheet of formation.
Using dough sheet, without the use of pixel, there are two benefits: firstly, reducing the matching error for forcing space continuous coupling；
Secondly, taking the average value of similar dough sheet when image denoising rather than similar pixel, several dough sheets belonging to each pixel can be obtained
To different estimated values.Its average value is taken again, and this operation commonly known as polymerize estimation (Aggregation of
Estimates), this can generate sizable optimization to the performance of image denoising.The dough sheet taken in the present invention is smaller, and (3 × 3 is big
It is small), historical relic model is received, the image rendered usually has good self-similarity (textural characteristics are similar), therefore meeting
There are many similar dough sheets.
Image denoising algorithm allows for estimated noise variance to calculate the similarity of noisy sample, in Monte Carlo wash with watercolours
In dye, each is estimated from the sample distribution that the irradiant average value of pixel and variance can be obtained directly from
Out.
To a pixel x, definitionFor meet the following conditions pixel y set: its center dough sheet PyMeetK is the standard for judging to have similar features dough sheet with pixel x, the maximum possible of noise-free pixel color
Estimator is exactly their arithmetic mean of instantaneous value:
Unlike the estimator for only calculating center dough sheet average value from front, the present invention can carry out going for monolith dough sheet
It makes an uproar, and then image can be denoised as unit of dough sheet.This be it is a kind of it is very classical by dough sheet denoise based on image go
It makes an uproar means.
To a noisy dough sheet P centered on pixel xx, by all dough sheet (i.e. Chi- for meeting following formula
Square distance is less than the dough sheet V for k) being averaged to be denoisedx:
HereIt is to all in dough sheet PyThe valuation of upper pixel u.
Although being denoised in this way to all dough sheets, not all pixel.Due to each face
Piece includes (2w+1)2A pixel, then corresponding each pixel is by (2w+1)2A dough sheet included, thus can obtain it is a large amount of right
The estimated value of its color finally can polymerize these estimated values at each pixel to generate final denoising image:
In formula, y-x is Chi-Square distance of the pixel y away from pixel x, and results are averaged to final, above to be exactly
The central principle of Noise Algorithm is removed as unit of dough sheet herein.
The realization of HM filter is very concise, in addition to calculating parameter required for light frequency curve, in the algorithm
It only needs to be arranged 4 parameters, is respectively as follows: the n of HM filter multiples, dough sheet half is w long, the b long and Chi- of search window half
The threshold value of Square distance.
The search of similar dough sheet is limited in the window that size is (2b+1) × (2b+1).Threshold value k is used to that mark is arranged
Quasi- Chi-Square distance.
The pseudocode that single overtones band curve merges HM filter is as follows:
Input: MC imageCorresponding frequency curve h, patch-sized w,
Search box size b, Chi-Square distance threshold k
Output: filtered image
The pseudocode that multiple frequence rate curve merges HM filter is as follows:
Input: MC imageCorresponding frequency curve h, patch-sized w, search box size b, Chi-Square distance
Threshold value k, filter multiple ns
Output: filtered image
It when starting rendering, is tentatively rendered first with less light sample size, the image rendered uses
The HM filter of threshold value k=0.5 carries out noise remove.
Step 2, the mean square error of current HM filter is estimated using SURE, and calculates each pixel in rendering image
Best light sample size.
The present invention provides a kind of iterative Adapti ve rendering method, core concept is using Stein ' s Unbiased
Risk Estimator (SURE) (to the unbiased estimating function of mean square error MSE (i.e. Mean Square Error)) determines most preferably to adopt
Sample density, i.e., best light sample size.It is using the benefit of SURE, can be used to instruct more by the mean square error that SURE is estimated
More sample distributions, therefore more samples will be assigned to nondescript region.
Monte Carlo ray tracing is by estimating the face of a pixel to sampling again reconstructed sample in integral domain at random
Color.Unbiased Monte Carlo Rendering have one can be by using the random error range that variance evaluation function estimates.
According to central-limit theorem, if Y is the pixel color estimated by unbiased Monte Carlo renderer, x is true
Pixel color, and number of samples n tends to be infinitely great, and it is x, variance σ that the distribution of Y, which is similar to an average value,2The normal state of/n
Distribution:
Herein: σ2It is the variance of Monte Carlo sample.For the sample of limited quantity, this relationship is also a kind of fine
Approximation.
Since Monte Carlo renderer is substantially a kind of estimation function, so carrying out estimation to its accuracy is very must
It wants.Stein ' s Unbiased Risk Estimator (SURE) gives a kind of one estimation function accuracy of estimation
Method.
SURE is pointed out, if y is in normal distributionX measurement, F is a weak differentiation function, then
Evaluated error are as follows:
To the unbiased esti-mator of the Mean Square Error of F (y), that is:
E [SURE (F (y))]=‖ F (y)-x ‖2
Above-mentioned formula shows: if σ can be calculatedyWith dF (y)/dy, estimation letter can be estimated not knowing the value of x
The error of number F.The present invention estimates optimal sample quantity required for every piece of region using above-mentioned formula.
It for each pixel, requires to determine the parameter in HM filter using minimum SURE error, to determine
The intensity of HM filter.
First calculate the dF (V of HM filter Fx)/dVx, it calculates as follows:
In formula: E2It (x) is evaluated error (E2(x) calculating uses the prior art, if the color value stability of pixel is good,
Then E2(x) value is smaller).
W (x, y) is weight of the pixel y relative to pixel x, represents x with p, represents y with q, be expressed as
Wherein: k is assignable decay factor, for controlling the intensity of filter.The smaller filter of the value of k is more conservative.
P hereinpAnd PqP is respectively indicated, the dough sheet of half a length of t, σ centered on q2Indicate general variance, it can be according to image averaging PSNR
It is calculated.
Its derivation process is as follows:
The expression formula of HM filter of the present invention is as follows:
It enablesSo have:
Again willIt substitutes into, after abbreviation, obtains:
Herein:
Then the mean square error (i.e. Mean Square Error, MSE) of HM filter is calculated.For each pixel, mistake
Color after filter can all be used to update Pixel Information.
The SURE calculated using Monte Carlo sample usually has noisy as a result, this is because SURE is the unbiased of MSE
Estimation function, and itself also has variance.In order to reduce variance, number of samples can be increased or be filtered.For right
Efficiency is considered, it will usually the variance of SURE is reduced using filtering.
Specifically, before using SURE optimization, first estimation MSE is schemed using parameter setting more conservative HM filter
As carrying out pre-filtering.
The MSE estimated value obtained using SURE can be used as the feedback information of renderer；Sampling density should be with MSE
Estimated value is directly proportional.But due to MSE estimated value and imperfect, the present invention is ensured by increasing a heuristic variance item
Region with higher variance can be assigned to more sample sizes.
In addition, it is more sensitive to dark area relative to bright areas human eye, in order to allow more light samples to be assigned to
Dark area, by sampling function multiplied by filtering pixel color of light according to the inverse square of value, to sum up, the sampling function S at pixel x
(x) are as follows:
Herein: E2It (x) is evaluated error；I(F(Vi))2For the color value of pixel after filtering；∈, which is one, prevents the divisor from being
Zero minimum, for example, being set as 0.001.If current sampling budget is m, received current best light at pixel x
Sample size is
By inputting the quantity of the threshold value k control removal noise of HM filter, the i.e. smoothness of control rendering image.Wash with watercolours
The optimum efficiency of dye depends on the light sample size needed in render process, is estimated at each pixel in this step using SURE
Required best light sample size, and optimal HM filter threshold value is obtained with this by calculating picture noise quantity.
Step 3, according to the amount of noise in rendering image, the threshold value of HM filter is updated.
Step 4, rendering is re-started to image according to best light sample size, and uses updated HM filter threshold
Value carries out noise remove.
Step 5, step 2~step 4 is repeated, until rendering image is met the requirements.
Picture quality is measured in the present invention using PSNR, the calculating formula of PSNR is as follows:
Herein:
Wherein, M is the length of image, and N is the width of image, fijFor the pixel of present image,For last iterative image
Pixel.
The present invention measures the variation of iterative image using MSE, and MSE can subtract with the increase of the number of iterations under normal circumstances
Few, corresponding PSNR will increase.When the rate that PSNR increases is slack-off, that is, PSNR shows when converging on some value, it can
To assert that current iteration income declines, the threshold value that iteration income is less than setting means that the quality of present image has tended to be full
With can terminate iterative process, obtain final image.
Iterative Adapti ve rendering pseudocode based on SURE optimization is as follows:
Input: original light sample size n0(being preset as 8spp), the threshold value k (being preset as 0.5) of HM filter, iteration are whole
Only threshold value(being preset as 0.1)
Output: result images u is generatedres
The considerations of for render time, if original light sample size n0=8, to the higher feelings of image quality requirements
Under condition, original light sample size can also be set higher.
The threshold value k size of HM filter is related with the HM filter removal intensity of noise, and the value of k is smaller, and HM filter is got over
Conservative, the quantity for removing noise can relatively low, but the image detail retained is higher, and when the value of threshold value k is bigger, HM filter removes
The ability of noise can become by force, and the details quality of corresponding image can decline.
Iteration ends threshold valueIndicate that the PSNR of current iteration image compares the change of image PSNR when last iteration is completed
Change amount,Smaller demand of the expression to picture quality of value it is higher, the number of iterations is also more, when the image income of grey iterative generation
It is less thanWhen, iterative process terminates, and obtains final image ures, the Δ PSNR=1 when carrying out first time iteration.
Claims (5)
1. a kind of iterative Adapti ve rendering method for merging HM filter using light frequency curve characterized by comprising
Step 1, image is tentatively rendered using less light sample size, is made an uproar using HM filter to rendering image
Sound removal；
Step 2, the mean square error of current HM filter is estimated using SURE, and is calculated in rendering image each using following formula
The best light sample size a of pixel:
In formula: m is the budget of light sample size；
S (x) is the light sampling function at pixel x；
S (y) is the light sampling function at pixel y；
For the set for meeting the following conditions pixel y:
Dough sheet P centered on pixel yyMeet
In formula:For the dough sheet P centered on pixel xxWith the dough sheet P centered on pixel yyDistance；
K is the threshold value of HM filter；
Step 3, according to the amount of noise in rendering image, the threshold value of HM filter is updated；
Step 4, rendering re-started to image according to best light sample size, and using updated HM filter threshold value into
Row noise remove；
Step 5, step 2~step 4 is repeated, until rendering image is met the requirements.
2. merge the iterative Adapti ve rendering method of HM filter using light frequency curve as described in claim 1,
It is characterized in that, the HM filter measures the similarity degree of pixel using the frequency curve of light sample at pixel.
3. merge the iterative Adapti ve rendering method of HM filter using light frequency curve as claimed in claim 2,
It is characterized in that, as unit of dough sheet, noise remove is carried out to rendering image using following formula:
In formula:For the color value for rendering pixel x in image after removal noise；
W is half length of dough sheet put centered on pixel x；
VyFor the result after the dough sheet denoising centered on pixel y；
Y-x is distance of the pixel y away from pixel x.
4. merge the iterative Adapti ve rendering method of HM filter using light frequency curve as claimed in claim 3,
It is characterized in that, the calculation formula of light sampling function S (x) is as follows:
In formula: VxFor the result after the dough sheet denoising centered on pixel x；
F(Vx) it is the filters filter function for filtering the dough sheet centered on pixel x；
SURE(F(Vx)) for using the mean square error of the SURE current HM filter calculated；
E2It (x) is the evaluated error of the color value of pixel x；
I(F(Vx))2It is the color value of pixel x after filtering；
∈, which is one, prevents the minimum numerical constant that divisor is zero.
5. merge the iterative Adapti ve rendering method of HM filter using light frequency curve as claimed in claim 4,
Be characterized in that, using PSNR measure rendering picture quality, set iteration ends threshold value asCurrent rendering image is compared to the last time
The PSNR variable quantity of rendering image is less than when iteration is completedWhen, iteration ends.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710054403.8A CN106898038B (en) | 2017-01-22 | 2017-01-22 | A kind of rendering method merging HM filter using light frequency curve |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710054403.8A CN106898038B (en) | 2017-01-22 | 2017-01-22 | A kind of rendering method merging HM filter using light frequency curve |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106898038A CN106898038A (en) | 2017-06-27 |
| CN106898038B true CN106898038B (en) | 2019-10-01 |
Family
ID=59199113
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710054403.8A Expired - Fee Related CN106898038B (en) | 2017-01-22 | 2017-01-22 | A kind of rendering method merging HM filter using light frequency curve |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106898038B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019182906A1 (en) | 2018-03-17 | 2019-09-26 | Nvidia Corporation | Shadow denoising in ray-tracing applications |
| US10991079B2 (en) * | 2018-08-14 | 2021-04-27 | Nvidia Corporation | Using previously rendered scene frames to reduce pixel noise |
| CN110033511A (en) * | 2019-04-19 | 2019-07-19 | 山东大学 | The predictor method of region unit computation complexity, parallel optical path method for tracing and system |
| CN112419492B (en) * | 2020-12-14 | 2022-08-23 | 长春理工大学 | Adaptive control method for sampling number of pixel path in visual perception driven Monte card rendering |
| CN116266375A (en) * | 2021-12-18 | 2023-06-20 | 华为技术有限公司 | Method, apparatus, device and storage medium for processing light ray data |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105513120A (en) * | 2015-12-11 | 2016-04-20 | 浙江传媒学院 | Adaptive rendering method based on weight local regression |
| CN105678831A (en) * | 2015-12-30 | 2016-06-15 | 北京奇艺世纪科技有限公司 | Image rendering method and apparatus |
| CN105893972A (en) * | 2016-04-08 | 2016-08-24 | 深圳市智绘科技有限公司 | Automatic illegal building monitoring method based on image and realization system thereof |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160098820A1 (en) * | 2014-10-03 | 2016-04-07 | Raghu Kopalle | System for robust denoising of images |
-
2017
- 2017-01-22 CN CN201710054403.8A patent/CN106898038B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105513120A (en) * | 2015-12-11 | 2016-04-20 | 浙江传媒学院 | Adaptive rendering method based on weight local regression |
| CN105678831A (en) * | 2015-12-30 | 2016-06-15 | 北京奇艺世纪科技有限公司 | Image rendering method and apparatus |
| CN105893972A (en) * | 2016-04-08 | 2016-08-24 | 深圳市智绘科技有限公司 | Automatic illegal building monitoring method based on image and realization system thereof |
Non-Patent Citations (2)
| Title |
|---|
| The SURE-LET approach to image denoising;Thierry Blu;《IEEE TRANSACTIONS ON IMAGE PROCESSING》;20071231;第16卷(第11期);第2778-2786页 * |
| 虚拟现实光线跟踪加速方法研究;王彦成;《科技与创新》;20160725(第14期);第110-111页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106898038A (en) | 2017-06-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106898038B (en) | A kind of rendering method merging HM filter using light frequency curve | |
| Wang et al. | Noise detection and image denoising based on fractional calculus | |
| CN103942758B (en) | Dark channel prior image dehazing method based on multiscale fusion | |
| Monteil et al. | A new interpretation and improvement of the nonlinear anisotropic diffusion for image enhancement | |
| Gao et al. | Sand-dust image restoration based on reversing the blue channel prior | |
| CN103116875B (en) | Self-adaptation bilateral filtering image de-noising method | |
| CN103679173B (en) | Method for detecting image salient region | |
| Wong | Adaptive bilateral filtering of image signals using local phase characteristics | |
| CN102800063A (en) | Image enhancement and abstraction method based on anisotropic filtering | |
| CN102708550A (en) | Blind deblurring algorithm based on natural image statistic property | |
| CN107292842A (en) | The image deblurring method suppressed based on prior-constrained and outlier | |
| CN108681995A (en) | A method of motion blur is gone based on variation Bayesian Estimation | |
| Bao et al. | An edge-preserving filtering framework for visibility restoration | |
| Raveendran et al. | Image fusion using LEP filtering and bilinear interpolation | |
| Yadav et al. | A partial differential equation-based general framework adapted to Rayleigh's, Rician's and Gaussian's distributed noise for restoration and enhancement of magnetic resonance image | |
| Du et al. | A new model for image segmentation | |
| Yang et al. | Hierarchical joint bilateral filtering for depth post-processing | |
| Sun et al. | Adaptive bilateral filter considering local characteristics | |
| Elad | Analysis of the bilateral filter | |
| Tang et al. | Single image dehazing algorithm based on sky segmentation | |
| Liu et al. | Adaptive level set image segmentation using the Mumford and Shah functional | |
| Huang et al. | Three-view dense disparity estimation with occlusion detection | |
| Jones | Feature preserving smoothing of 3D surface scans | |
| Kaur et al. | Comparison of noise removal techniques using bilateral filter | |
| Liu et al. | Improved block Kalman filter for degraded image restoration |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191001 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |