CN102654914A - Method for accelerating image haze removal by utilizing image processing unit - Google Patents
Method for accelerating image haze removal by utilizing image processing unit Download PDFInfo
- Publication number
- CN102654914A CN102654914A CN2011100571502A CN201110057150A CN102654914A CN 102654914 A CN102654914 A CN 102654914A CN 2011100571502 A CN2011100571502 A CN 2011100571502A CN 201110057150 A CN201110057150 A CN 201110057150A CN 102654914 A CN102654914 A CN 102654914A
- Authority
- CN
- China
- Prior art keywords
- value
- passage
- pixel
- image
- gray
- 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
Images
Abstract
The invention discloses a method for accelerating image haze removal by utilizing an image processing unit. The method comprises the following steps of: obtaining a minimum gray value of a local pixel block by taking each pixel of a hazed image as the center; calculating average values of gray values of R, G and B channels of all pixels of a region, covered by a preset brightest region in a dark primary-color image, in a hazed image, and taking the maximum value of the average values as an atmospheric optical value of the hazed image; carrying out normalization on a minimum gray value by utilizing the atmospheric optical value, and obtaining a rough transmission graph by utilizing a normalization result; constructing a laplace matrix according to the gray values of the R, G and B channels of the all pixels in the hazed image, and optimizing the rough transmission graph by utilizing the laplace matrix; and obtaining the gray values of the R, G and B channels of each pixel of the hazed image by utilizing the optimal transmission graph, the atmospheric optical value, and the gray values of the R, G and B channels of each pixel in the hazed image.
Description
Technical field
The present invention relates to the graphics process field, relate more specifically to the method for a kind of use GPU accelerogram as mist elimination (image defogging).
Background technology
Under the relatively poor situation of weather condition, the sharpness of image and color usually can be by the fog deteriorations in the atmosphere.Fair need the processing through mist elimination of image that in this weather, captures and video improved.The process of removing the fog effect in the image is called as the image mist elimination.The image mist elimination is for the navigation of under the relatively poor situation of weather condition, carrying out and keep watch on very useful.
The a lot of image defogging method capables of current existence are best a kind of of effect based on the image defogging method capable of dark primary priori (dark channel prior) wherein.Dark primary priori draws through the no mist image data base in open air is added up, and, all has such pixel in each regional area of most no mist images in open air that is, and the gray-scale value of at least one Color Channel of this pixel is very low.The mist elimination model that utilizes dark primary priori to set up can directly be estimated the concentration of fog and can be the image (abbreviating the mist elimination image as) after high-quality removal fog disturbs with mist image restoration is arranged.
In image defogging method capable, through utilizing the gray-scale value I that the mist image is arranged, atmosphere light value A and the transmission plot t of input, according to there being mist iconic model I=Jt+A (1-t) to solve the gray-scale value J of mist elimination image based on dark primary priori.Process based on the image defogging method capable of dark primary priori is simple relatively, but the travelling speed on CPU (CPU) is very slow.For example, on the personal computer of 4 processors of the intel pentium with 3.0GHz, the image of handling 600 * 400 pixels approximately need spend 10 to 20 seconds time.So, in using in real time, unlikely use this image defogging method capable.
Summary of the invention
Problem in view of the above, what the present invention proposes a kind of novelty utilizes the method for GPU accelerogram as mist elimination.
Image defogging method capable according to an embodiment of the invention comprises: obtaining so that each pixel in the mist image to be arranged is the minimum gradation value of the local pixel piece at center; Be dark primary, and utilize corresponding to the dark primary structure dark primary image that each pixel in the mist image is arranged; The mean value of the gray-scale value of the mean value of the gray-scale value of the mean value of brightest area gray-scale value of the R passage of all pixels of institute overlay area in the mist image is arranged of predetermined size, G passage and B passage in the calculated dark primary colour image, and with the maximal value in the said mean value as the atmosphere light value that the mist image is arranged; Utilize the atmosphere light value that minimum gradation value is carried out normalization, and utilize the normalization result to obtain rough transmission plot; According to the gray-scale value structure Laplce matrix of the R passage that the pixel in the mist image is arranged, G passage and B passage, and utilize Laplce's matrix that rough transmission plot is optimized; Utilize transmission plot, atmosphere light value of optimizing and the gray-scale value that R passage, G passage and the B passage of each pixel in the mist image are arranged, obtain the gray-scale value of R passage, G passage and the B passage of each pixel in the mist elimination image.
In the method that proposes, more than all processes all realize by GPU GPU.The present invention is intended to convert the CPU execution flow process of serial into parallel GPU execution flow process, thereby improves travelling speed greatly.
Description of drawings
From below in conjunction with understanding the present invention better the description of accompanying drawing specific embodiments of the invention, wherein:
Fig. 1 shows the synoptic diagram of the processing streamline of GPU;
Fig. 2 shows the entire process process based on the image defogging method capable of dark primary priori according to the embodiment of the invention.
Embodiment
To describe the characteristic and the exemplary embodiment of various aspects of the present invention below in detail.Many details have been contained in following description, so that complete understanding of the present invention is provided.But, it will be apparent to one skilled in the art that the present invention can implement under the situation of some details in not needing these details.Description in the face of embodiment only is in order through example of the present invention is shown the clearer understanding to the present invention to be provided down.Any concrete configuration and the algorithm that are proposed below the present invention never is limited to, but any modification, replacement and the improvement that under the prerequisite that does not break away from spirit of the present invention, have covered coherent element, parts and algorithm.
GPU (GPU) has powerful computation capability, and it can also be used to carry out the calculating (that is, being embodied as GPUPU) of general purpose except can being used for traditional graphical application.The present invention attempts the image defogging method capable based on dark primary priori is mapped in the computing environment based on the GPU of OpenGL ES2.0 standard, to realize the image defogging method capable based on dark primary priori in real time.
Traditional image defogging method capable based on dark primary priori comprises 5 steps: step 1, and according to equality I
Dark=min
Ω(min
C(I
C)); Obtain rough dark primary from the gray-scale value I that the mist image is arranged, wherein Ω is to be the local pixel piece (for example, 15 * 15 block of pixels) at center with the current pixel; C be Color Channel (for example; R passage, G passage or B passage) expression (that is, find out current pixel and be the gray-scale value of the minimum Color Channel of gray-scale value among the R, G, B passage of the pixel (they have formed local pixel piece Ω) at center, the i.e. minimum gradation value of local pixel piece Ω) with it; Step 2; Find out in the dark primary image 0.1% brightest area; Find out the gray-scale value of the Color Channel that gray-scale value is the highest among the R, G, B passage of this brightest area all pixels of institute overlay area in the mist image is arranged then; Be the highest gray-scale value of brightest area, and this highest gray-scale value is set to the atmosphere light value A of each Color Channel of each pixel in the mist image; Step 3, the calculating normalization dark primary
Basis then
Calculate rough transmission plot
Wherein ω is weighting factor (being generally 0.95); Step 4, the transmission plot t that uses the stingy figure of soft picture (image soft matting) to obtain optimization (particularly, uses soft picture to scratch figure and finds the solution sparse linear systems
Wherein L is from Laplce's matrix of mist picture construction is arranged, and λ is a little value parameter (for example, 10
-4), U is a unit matrix; Step 5 recovers mist elimination image (that is, drawing the gray-scale value of each Color Channel of each pixel in the mist elimination image) according to I=Jt+A (1-t), and wherein, J representes the gray-scale value of R, G or the B passage of the pixel in the mist elimination image, J=(I-A)/max (t, t
0)+A, t
0Be threshold parameter (being generally 0.1).In above step, step 4 is the most complicated and need carry out iteration.
For what comprise the individual pixel of M * N for example (M capable * M row) the mist image arranged; Through step 1 can draw with this have each pixel in the mist image be the center the local pixel piece minimum gradation value (promptly; Can draw M * N minimum gradation value); Can draw the maximum gradation value that this has in the mist image 0.1% brightest area through step 2; Can draw size respectively through step 3 and step 4 and be the rough transmission plot of M * N and the transmission plot of size, can utilize minimum gradation value that step 1 draws in 4, maximum gradation value, rough transmission plot then and the transmission plot optimized draws the gray-scale value through each Color Channel after the mist elimination processing of M * N pixel through step 5 for the optimization of M * N.
Standard OpenGL ES2.0 be a kind of global function two dimension on the embedded system with three-dimensional picture exempt from permit, cross-platform graphics application program interface (API).The last application of CPU can be controlled GPU through calling API.Fig. 1 shows the processing streamline of GPU.In the processing streamline of GPU, vertex shader and fragment shader are two programmable stages, and comprise that it is stages of fixing that pel assembling, raster scanning and every segment operate in other interior stages.
Vertex shader realizes being used for the summit is carried out the general programmable method of computing.Wherein, the summit is the end points on line segment, triangle or the polygon, and the computing on summit is comprised evolution, color calculation and texture coordinate generation etc.Pel is the geometric object such as line segment, triangle and polygon etc.At the pel assembling stage, the summit will be mounted to each pel, and GPU will carry out cutting and rejecting under possible situation.In the raster scanning stage, pel will be converted into segment.Segment is represented the pixel of drawing on the display.Fragment shader realizes being used for segment is carried out the general programmable method of computing.In every segment operational phase, GPU will carry out the test of pixel ownership, intercepting test, stencil test, depth test, colour mixture and shake demonstration etc.The GPU that follows OpenGL ES2.0 standard is owing to comprising that the stage able to programme is called as Programmable GPU.A complete GPU handles streamline and is called as a render process.Because fragment shader has for the more complicated visit of texture storage device and has more computational resource, so will mainly be realized by fragment shader based on the image defogging method capable of dark primary priori.
In one embodiment of the invention, GPU will realize the image defogging method capable based on dark primary priori through a plurality of render process.Wherein, in all render process, vertex shader is used to obtain the interior location coordinate and the inner vein coordinate on each summit of mist image (image that perhaps draws through one or more render process).Particularly, vertex shader has following form:
Vertex?shader
Wherein, a_position be the summit of rectangle position coordinates (x, y, z); Vec () is vector (x, y, the z that combines a_position and 1.0 to draw; 1.0), gl_Position is the interior location coordinate of OpenGL ES background (that is, among the GPU); A_texCoord is the texture coordinate on the summit of rectangle, and v_texCoord is the inner vein coordinate of OpenGL Es background (that is, among the GPU).Fragment shader will use v_texCoord to come in the texture of R, G, B passage gray-scale value, the gray-scale value of G passage and the gray-scale value of B passage of the R passage of each pixel are sampled.
In order to be mapped to based on the image defogging method capable of dark primary priori in the computing environment based on the GPU of OpenGL ES 2.0, also carry out 5 steps according to one embodiment of present invention, each step is accomplished through one or more render process.Fig. 2 shows based on the entire process process of the image defogging method capable of dark primary priori (tinter among Fig. 2 is realized by fragment shader).
Before carrying out processing procedure shown in Figure 2; The mist image (size is M * N pixel) that has that is input to GPU from CPU is divided into 3 Color Channels (promptly; R, G, B passage) texture (size of each texture is M * N), and is stored in the graphic memory.The texture of these 3 passages will be used in steps in institute shown in Figure 2.Having the mist image will be used as the rectangle that is of a size of M * N plays up.
In step S202, fragment shader is according to equality I
Dark=min
Ω(min
C(I
C)) obtain rough dark primary I
Dark, and with rough dark primary I
DarkDeposit in (when carrying out this step, fragment shader is called as the dark primary tinter) in the frame buffer with texture object.Step S202 is accomplished by a render process.In according to one embodiment of present invention, be that the size of the local pixel piece Ω at center is set to 15 * 15 with the current pixel.In step S202, the dark primary tinter is searched for the minimum gradation value in R, G, three Color Channels of B respectively in local pixel piece Ω
Select then
A gray-scale value of middle minimum is as the I of local pixel piece Ω
DarkThat is to say that it is the minimum gradation value of the local pixel piece at center that the dark primary tinter can obtain so that each pixel in the mist image to be arranged.The dark primary tinter has following form:
Dark?channel?fragment?shader
Wherein, v_texCoord is the inner vein coordinate from the summit of the rectangle of vertex shader, and texture2D () is the function of sampling 2 d texture, and gl_FragColor is with the segment color value that is rendered into frame buffer.Minimum gradation value will be deposited in the frame buffer.
In step S204; Predetermined ratio in the fragment shader calculated dark primary colour image (for example; The mean value of the mean value of the mean value of the gray-scale value of the R passage of all pixels of institute overlay area in the mist image is arranged of brightest area 0.1%), the gray-scale value of G passage and the gray-scale value of B passage (when carrying out this step, fragment shader is called as atmosphere optical colour device).Step S204 is accomplished by a render process.Wherein, the size value of this brightest area (for example, H capable * V row) sends to GPU as constant by CPU.Atmosphere optical colour device is set to R, the G of the center pixel of brightest area, the gray-scale value of B passage respectively with the mean value of the gray-scale value of the mean value of the gray-scale value of the mean value of the gray-scale value of the R passage of all pixels in the brightest area, G passage and B passage, and the gray-scale value of the R of this center pixel of brightest area, G, B passage is deposited in the frame buffer.Then, CPU finds out the maximal value in the gray-scale value of R, G, B passage of this center pixel of brightest area, and A is sent to GPU as the atmosphere light value.Atmosphere optical colour utensil has following form:
Air?light?fragment?shader
Wherein, vec2 () utilizes 2 elements to make up the function of vector.
In walking rapid S206; Fragment shader is according to equality
calculating normalization dark primary; Rough according to equality
calculating then transmission plot
(when execution in step, fragment shader is called as the transmission plot tinter).Step S206 is accomplished by a render process.The transmission plot tinter has following form:
Transmission?map?fragment?shader
Wherein, omega (being generally 0.95) and atmosphere light value A are constants, and will be deposited in the frame buffer by the result of calculation that CPU sends GPU transmission plot tinter to.
In step S208, fragment shader is optimized rough transmission plot.Step S208 is accomplished by four render process.Wherein, two render process are used to make up Laplce's matrix, and latter two render process is used to find the solution sparse linear systems.Three render process are complete render process, and last render process is unfixed (because iteration depends on initial value and concrete image).When execution in step S208, fragment shader is called as transmission plot and optimizes tinter.
Laplce's matrix element can draw according to following equality:
Wherein, δ
I, jBe Kronecker function (Kronecker delta) (when i=j, δ
I, j=1; When i ≠ j, δ
I, j=0), μ
kAnd C
kBe the window w that slides on the mist image having
kThe mean value matrix and the covariance matrix of the gray-scale value of the R of the pixel in (3 * 3 window), G, B passage, U
3Be 3 * 3 unit matrixs, α is that value is 10
-4/ 9 parameter, I
iAnd I
jBe window w
kThe gray-scale value (I that piles up R that the address is the pixel of i and j, G, B passage in covered 3 * 3
iAnd I
jBe the matrix that comprises the gray-scale value of R, G, three passages of B), i, j, k are the addresses of piling up by row that the pixel in the mist image is arranged, and change in the scope of M * N 1.(i, what j) describe is the dependence that on some feature space, exists between the image local pixel to L.
The processing that makes up Laplce's matrix generally includes following two processing (that is to say, transmission plot optimization tinter through two render process from mist picture construction Laplce matrix is arranged):
At first, fragment shader is calculated l through a render process according to following equality
I, j, the fragment shader of this moment is called as Laplce's matrix element tinter:
Laplce's matrix element tinter has following form:
Laplacian?matrix?element?fragment?shader
Wherein, pack is used for the pixel value packing to 3 * 3, and mean is μ
k, cov is C
k, pack.i and pack.j be meant all pixel values in 3 * 3 sampled, result_pack is the data structure that the value packing that pixel i and j share is drawn and is stored in the frame buffer.
Then, fragment shader through render process according to following equality calculate L (i, j), the fragment shader of this moment is called as Laplce's matrix summation tinter:
Laplce's matrix summation tinter has following form:
Laplacian?matrix?summation?fragment?shader
Wherein, texCoord is used for storage l
I, jTexture in all coherent elements texture coordinate of sampling.Function sum () will sue for peace to coherent element, and the result will be stored in the frame buffer.
Conjugate gradient (CG) algorithm is used to find the solution sparse linear systems.Fragment shader is through carrying out iterative processing, draws the transmission plot after the optimization.
At first, fragment shader is played up the cycle basis through one
Draw initial iteration intermediate variable d
0And r
0At this moment, fragment shader is called as solver initialization tinter, has following form:
Solver?initialization?fragment?shader
Wherein, lap_row comes out from Laplce's matrix texture sampling, and tc comes out from rough transmission plot (transmission plot that step S206 draws) texture sampling, and t0 is initialization transmission plot texture and sends GPU to by CPU.T0 can get some certain fixed value between 0.0 to 1.0 (such as 0.0,0.5,1.0 or other values) simply, but in order to accelerate iterative convergence speed, generally can use prerequisite (Preconditioning) to carry out initialization.After calculating, the result will be packaged in the texture of transmission plot, and be stored in the frame buffer.
Then, fragment shader is carried out iterative processing according to following equality, up to d
iAnd r
iStop iteration till reaching the threshold value of expectation, and obtain the transmission plot after the final optimization.
t
i+1＝t
i+α
id
i，
r
i+1＝r
i-α
iLd
i，
d
i+1＝r
i+1+β
i+1d
i
When carrying out iterative processing, fragment shader is called as conjugate gradient solver tinter, has following form:
Conjugate?gradient?solver?fragment?shader
Wherein, pack_t.r, pack_t.d, pack_t.t are the components from pack_t, and pack () is the function that some compositions is packaged as a texture.
In step S210, fragment shader is according to equality J=(I-A)/max (t, t
0)+A recovers mist elimination image (that is, drawing R, the G of each pixel in the mist elimination image, the gray-scale value of B passage).Step S210 is accomplished by a render process.When execution in step S210, fragment shader is called as the recovery tinter, has following form:
Restoration?fragment?shader
Wherein, t0 is the constant that sends to GPU from CPU, and output is R, the G of each pixel in the mist elimination image, the gray-scale value of B passage.
With respect to the image defogging method capable of on CPU, realizing based on dark primary priori; The present invention has the following advantages: 1) because GPU can be concurrently to there being each pixel on the mist image to carry out the processing of image mist elimination, so implementation of the present invention is faster than the implementation on the CPU.2), be easy to use hardware to realize so compare other GPU general-purpose computations high-level programming languages because the present invention is based on the tinter from OpenGL ES 2.0.
Below the present invention has been described with reference to specific embodiment of the present invention; But those skilled in the art all understand; Can carry out various modifications, combination and change to these specific embodiments, and can not break away from the spirit and scope of the present invention that limit accompanying claims or its equivalent.
Can come execution in step with hardware or software as required.Notice that without departing from the scope of the invention, the process flow diagram that can in this instructions, provide adds step, therefrom removes step or revise step wherein.In general, process flow diagram just is used for indicating a kind of possible sequence of the basic operation that is used to realize function.
The universal digital computer of embodiments of the invention programming capable of using, utilize special IC, PLD, field programmable gate array, light, chemistry, biological, system quantum or nanometer engineering, assembly and mechanism to realize.In general, function of the present invention can be realized by any means known in the art.Can use distributed or networked system, assembly and circuit.The communication of data or to transmit can be wired, wireless or through any other means.
Also will recognize, according to the needs of application-specific, one or more can perhaps even in some cases being removed or being deactivated in the key element shown in the accompanying drawing by more separating or more integrated mode realizes.Program or code that realization can be stored in the machine readable media are carried out above-mentioned any method to allow computing machine, also within the spirit and scope of the present invention.
In addition, it only is exemplary that any signal arrows in the accompanying drawing should be considered to, rather than restrictive, only if concrete indication is arranged in addition.Separate or the ability of combination when not knowing when term is also contemplated as to make, the combination of assembly or step also will be considered to put down in writing.
Claims (7)
1. method of using the GPU accelerogram as mist elimination comprises following process:
Obtaining so that each pixel in the mist image to be arranged is the minimum gradation value of the local pixel piece at center, i.e. dark primary, and utilization has the dark primary of each pixel in the mist image to make up the dark primary image corresponding to said;
The brightest area of calculating in the said dark primary image predetermined size is at the mean value of the gray-scale value of the mean value of the gray-scale value of the said mean value that the gray-scale value of the R passage of all pixels of institute overlay area in the mist image arranged, G passage and B passage, and with the maximal value in the said mean value as the said atmosphere light value that the mist image is arranged;
Utilize said atmosphere light value that said minimum gradation value is carried out normalization, and utilize the normalization result to obtain rough transmission plot;
According to the said gray-scale value structure Laplce matrix that R passage, G passage and the B passage of the pixel in the mist image are arranged, and utilize said Laplce's matrix that said rough transmission plot is optimized; And
Utilize transmission plot, said atmosphere light value and the said gray-scale value that R passage, G passage and the B passage of each pixel in the mist image are arranged optimized, obtain the gray-scale value of R passage, G passage and the B passage of each pixel in the mist elimination image.
2. use GPU accelerogram according to claim 1 is as the method for mist elimination; It is characterized in that; Said method is by GPU and the collaborative completion of CPU, and wherein said CPU is as main control unit, and said GPU is as quickening arithmetic element.
3. use GPU accelerogram according to claim 2 is as the method for mist elimination; It is characterized in that; Fragment shader in the said GPU is calculated the mean value of said brightest area at the gray-scale value of the mean value of the gray-scale value of the said mean value that the gray-scale value of the R passage of all pixels of institute overlay area in the mist image arranged, G passage and B passage, and the maximal value that said CPU is found out in the said mean value sends said GPU to as the said atmosphere light value that the mist image is arranged.
4. image defogging method capable according to claim 2 is characterized in that the fragment shader in the said GPU is according to equality I
Dark=min
Ω(min
C(I
C)), obtain with said that each pixel in the mist image is arranged is the minimum gradation value of the local pixel piece at center, wherein Ω representes any local pixel piece that pixel pixel j is the center of having in the mist image with said, I
DarkThe minimum gradation value of representing said local pixel piece Ω, I
CRepresent the gray-scale value of R passage, G passage or the B passage of said pixel j, min () is the function of minimizing, and j is the integer greater than 1.
5. use GPU accelerogram according to claim 2 is characterized in that as the method for mist elimination the fragment shader in the said GPU is according to equality
Said minimum gradation value is carried out normalization, and according to equality
Obtain said rough transmission plot, wherein Ω representes with the said local pixel piece that any pixel j in the mist image is arranged is the center,
The normalization result who representes the minimum gradation value of said local pixel piece Ω, I
CThe gray-scale value of representing R passage, G passage or the B passage of said pixel j,
Expression is corresponding to the rough transmission plot of said pixel j, and A representes said atmosphere light value, and ω is first predetermined value that is sent to said GPU by said CPU, and min () is the function of minimizing, and j is the integer greater than 1.
6. use GPU accelerogram according to claim 2 is characterized in that as the method for mist elimination the fragment shader in the said GPU is constructed Laplce's matrix according to following equality,
Wherein, L (i, the j) element that i is capable, j is listed as in the said Laplce's matrix of expression, δ
I, jBe the Kronecker function, μ
kAnd C
kBe 3 * 3 window w that slide on the mist image to be arranged said
kIn the mean value matrix and the covariance matrix of gray-scale value of R, G, B passage of pixel, U
3Be 3 * 3 unit matrixs that sent to said GPU by said CPU, α is that the value that sends said GPU to by said CPU is 10
-4/ 9 parameter, I
iAnd I
jBe said window w
kThe matrix that the gray-scale value that piles up R that the address is the pixel of i and j, G, B passage in covered 3 * 3 is formed, i, j, k are the said addresses of piling up by row that pixel in the mist image is arranged.
7. use GPU accelerogram according to claim 5 is as the method for mist elimination; It is characterized in that said fragment shader is utilized said Laplce's matrix according to following equality and sent to the second predetermined value t of said GPU by said CPU
0Come the transmission plot t after iteration draws said optimization:
r
i+1＝r
i-α
iLd
i，，
d
i+1＝r
i+1+β
i+1d
i
Wherein, said fragment shader is at r
iSatisfy termination of iterations under the situation of predetermined threshold requirement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011100571502A CN102654914A (en) | 2011-03-04 | 2011-03-04 | Method for accelerating image haze removal by utilizing image processing unit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011100571502A CN102654914A (en) | 2011-03-04 | 2011-03-04 | Method for accelerating image haze removal by utilizing image processing unit |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102654914A true CN102654914A (en) | 2012-09-05 |
Family
ID=46730540
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011100571502A Pending CN102654914A (en) | 2011-03-04 | 2011-03-04 | Method for accelerating image haze removal by utilizing image processing unit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102654914A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103049890A (en) * | 2013-01-17 | 2013-04-17 | 四川川大智胜软件股份有限公司 | Real-time image defogging method based on CUDA (Compute Unified Device Architecture) |
| CN104881879A (en) * | 2015-06-15 | 2015-09-02 | 北京航空航天大学 | Remote sensing image haze simulation method based on dark-channel priori knowledge |
| CN105303524A (en) * | 2014-06-20 | 2016-02-03 | 现代自动车株式会社 | Apparatus and method for removing fog in image |
| CN106447639A (en) * | 2016-10-18 | 2017-02-22 | 乐视控股（北京）有限公司 | Mobile terminal photograph processing method and device |
| CN108010113A (en) * | 2017-11-21 | 2018-05-08 | 成都品果科技有限公司 | A kind of deep learning model based on pixel coloring device performs method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101783012A (en) * | 2010-04-06 | 2010-07-21 | 中南大学 | Automatic image defogging method based on dark primary colour |
| CN101882311A (en) * | 2010-06-08 | 2010-11-10 | 中国科学院自动化研究所 | Background modeling acceleration method based on CUDA (Compute Unified Device Architecture) technology |
| CN101950416A (en) * | 2010-09-15 | 2011-01-19 | 北京理工大学 | Bidirectional filtration-based real-time image de-hazing and enhancing method |
-
2011
- 2011-03-04 CN CN2011100571502A patent/CN102654914A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101783012A (en) * | 2010-04-06 | 2010-07-21 | 中南大学 | Automatic image defogging method based on dark primary colour |
| CN101882311A (en) * | 2010-06-08 | 2010-11-10 | 中国科学院自动化研究所 | Background modeling acceleration method based on CUDA (Compute Unified Device Architecture) technology |
| CN101950416A (en) * | 2010-09-15 | 2011-01-19 | 北京理工大学 | Bidirectional filtration-based real-time image de-hazing and enhancing method |
Non-Patent Citations (3)
| Title |
|---|
| KAIMING HE ET AL.: "Single image haze removal using dark channel prior", 《IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION 2009》 * |
| LENIN,A. ET AL.: "A Closed-Form Solution to Natural Image Matting", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 吴思华，柳有权: "基于图形处理器(GPU)的通用计算", 《计算机辅助设计与图形学学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103049890A (en) * | 2013-01-17 | 2013-04-17 | 四川川大智胜软件股份有限公司 | Real-time image defogging method based on CUDA (Compute Unified Device Architecture) |
| CN105303524A (en) * | 2014-06-20 | 2016-02-03 | 现代自动车株式会社 | Apparatus and method for removing fog in image |
| CN105303524B (en) * | 2014-06-20 | 2020-03-06 | 现代自动车株式会社 | Image defogging device and method |
| CN104881879A (en) * | 2015-06-15 | 2015-09-02 | 北京航空航天大学 | Remote sensing image haze simulation method based on dark-channel priori knowledge |
| CN104881879B (en) * | 2015-06-15 | 2017-10-13 | 北京航空航天大学 | A kind of remote sensing images haze emulation mode based on dark channel prior |
| CN106447639A (en) * | 2016-10-18 | 2017-02-22 | 乐视控股（北京）有限公司 | Mobile terminal photograph processing method and device |
| CN108010113A (en) * | 2017-11-21 | 2018-05-08 | 成都品果科技有限公司 | A kind of deep learning model based on pixel coloring device performs method |
| CN108010113B (en) * | 2017-11-21 | 2021-07-27 | 成都品果科技有限公司 | Deep learning model execution method based on pixel shader |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Steinbrücker et al. | Volumetric 3D mapping in real-time on a CPU | |
| US9508181B2 (en) | Ordering and rendering buffers for complex scenes with cyclic dependency | |
| EP3129974B1 (en) | Gradient adjustment for texture mapping to non-orthonormal grid | |
| EP3180773B1 (en) | Bandwidth reduction using texture lookup by adaptive shading | |
| JP4938850B2 (en) | Graphic processing unit with extended vertex cache | |
| CN111066066B (en) | Variable ratio tinting | |
| JP5451285B2 (en) | Image processing apparatus and image processing method | |
| CN100504923C (en) | Grain engine for grain video signal processing, graph processor and method | |
| US8072452B1 (en) | Efficient multi-resolution curve rendering | |
| CN103810728A (en) | System, method, and computer program product for inputting modified coverage data into a pixel shader | |
| CN102654914A (en) | Method for accelerating image haze removal by utilizing image processing unit | |
| JP2018512644A (en) | System and method for reducing memory bandwidth using low quality tiles | |
| CN106575429A (en) | High order filtering in a graphics processing unit | |
| TWI769138B (en) | Graphics processing method, pipeline circuit and computer program | |
| US7432937B2 (en) | System and method for concave polygon rasterization | |
| CN107038729B (en) | Digital instrument panel drawing method based on OpenGL-ES | |
| US7907147B2 (en) | Texture filtering apparatus, texture mapping apparatus, and method and program therefor | |
| CN106575428A (en) | High order filtering in a graphics processing unit | |
| US20150103252A1 (en) | System, method, and computer program product for gamma correction in a video or image processing engine | |
| EP4123106A1 (en) | Image processing device, image processing method, and image processing program | |
| US20150242988A1 (en) | Methods of eliminating redundant rendering of frames | |
| EP2728551B1 (en) | Image rendering method and system | |
| US11417058B2 (en) | Anti-aliasing two-dimensional vector graphics using a multi-vertex buffer | |
| CN111127297B (en) | Vector map solid line symbol drawing method considering line width consistency | |
| EP2702584B1 (en) | Analytical multi-view rasterization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20120905 |