CN104077745B - Image demisting device and method - Google Patents
Image demisting device and method Download PDFInfo
- Publication number
- CN104077745B CN104077745B CN201310106782.2A CN201310106782A CN104077745B CN 104077745 B CN104077745 B CN 104077745B CN 201310106782 A CN201310106782 A CN 201310106782A CN 104077745 B CN104077745 B CN 104077745B
- Authority
- CN
- China
- Prior art keywords
- distance parameter
- image
- thick distance
- current pixel
- sky
- 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
Abstract
The invention provides an image demisting device and method. The device comprises an input unit used for inputting an original image, a first rough distance parameter obtaining unit used for calculating the rough distance parameters of each pixel in the original image, an edge obtaining unit used for calculating the edge information of the original image, a sky separating unit used for separating the sky area in the original image based on the edge information and the rough distance parameters, a second rough distance parameter obtaining unit used for correcting the rough distance parameters according to a sky separating result output by the sky separating unit, a rough distance parameter refining unit used for refining the corrected rough distance parameters, and a processing unit used for conducting demisting and white balancing by means of the refined rough distance parameters obtained through the rough distance parameter refining unit. By the adoption of the image demisting device, feature information of the image can be fully utilized, a better demisting effect can be obtained for a dust scene or a colored mist scene, and image distortion is avoided.
Description
Technical field
The present invention relates to a kind of image demister and image haze removal method.
Background technology
Chinese geography environment is complicated, and weather is various, and the greasy weather is often visible in most of regional Jing, and indivedual areas can also be received
To the impact of dust and sand weather.Mist reduces the visibility of outdoor video image with dust and sand weather.In addition, in large size city, by
In the problem of air quality, the visibility of outdoor video image also can be reduced.The reduction of visibility is for the quality of video image
Produce considerable influence.Especially for protection and monitor field, affect to become apparent from.Atomization image sharpening is image processing field
In urgent functional need.Cause recent years to be atomized at image demisting based on the breakthrough of the demisting technology of single-frame images simultaneously
Reason becomes the more and more important research direction of computer vision field.
At present, being atomized image clarification method mainly has both direction：Demisting algorithm based on general pattern Enhancement Method
With the demisting algorithm based on Atmospheric models.Belong to has histogram average, sky based on the demisting algorithm of general pattern Enhancement Method
Between sharp filtering, high frequency enhancing filter, small echo strengthen, Retinex strengthen filtering etc..These algorithms are strengthening the contrast of image
Degree and visibility are starting point and are not directly dependent upon with demisting, for the video image caused by reasons such as low illuminations can be shown in
The problems such as degree is reduced can also be processed.In 2002, by NARASIMAHAN et al. in paper《Vision and the
Atmosphere》In propose first based on the demisting clarification method of Atmospheric models.This kind of method obtains skill in recent years
Art breaks through, and mainly by Fattal, Kaiming He et al. propose some new algorithms, in the situation of only single-frame images
Under, except fog effect is far superior to the enhanced defogging method of general pattern.
Atmospheric physics model describes the camera head when having suspended particulate in air and shoots or eye-observation object
Optical principle.The formula of Atmospheric models is：
I(X)=J(X)t(X)+A(1-t(X))
（1）
Wherein, I (X) represents the atomization image that camera head photographs or the atomization image that eye-observation is arrived, X=(x, y)
For image pixel coordinates.J (X) is object reflected light image, represents the image without mist, or the knot that can be described as demisting process
Fruit image.A be in image sky a bit（Below it is also referred to as " day null point "）Rgb value, also referred to as day null point parameter below.Such as
Exist without sky in fruit current input image, then the most strong point of mistiness degree in image is regarded as into a day null point.T (X) defines sky
The transfer function of gas medium, describes object reflected light and left behind after the scattering of airborne particles and reach shooting dress
The ratio put.T (X) is one and is more than 0 and the scalar data less than 1, and each pixel has a t (X) in image.I(X)、J
(X) it is image RGB data with A.
Referring to Fig. 1 formulas (1).Fig. 1 is the schematic diagram of Atmospheric models formula.Image on the left of Fig. 1 be human eye or
The image I (X) that camera head is observed.Image I (X) is made up of two parts, wherein, Part I is object reflected light J (X) Jing
Part J (X) t (X) remained after airborne particles scattering is crossed, Part II is that airborne particles scatter sunshine
Caused atmospheric environment light A (1-t (X)).T (X) in formula (1) is subject and camera head（Human eye）Between distance
（That is object distance）Function, be embodied as
t(X)=e-βd(X)（2）
Wherein, d (X) is an object point and camera head in image（Human eye）The distance between, thus t (X) be also referred to as " away from
From parameter ".β is atmospheric scattering coefficient, is constant.
By formula (1) with formula (2) it can be seen that object reflected light reaches intensity J (X) t (X) and object of camera head and takes the photograph
As the relation between the distance between device d (X) is that the more remote then light attenuation of distance is more severe；Atmospheric environment light reaches shooting dress
The intensity A (1-t (X)) put and the relation between d (X) are that the more remote then light of distance is stronger, so presenting at infinity
Go out white.
Recent years, made a breakthrough based on the demisting algorithm of Atmospheric models formula (1), these algorithms only need single-frame images to make
It is good except fog effect with regard to obtaining for input picture.Some related algorithms are listed in table 1.
Demisting algorithm of the table 1 based on Atmospheric models
Compared with traditional algorithm for image enhancement, these are obtained in that more preferable demisting based on the demisting algorithm of Atmospheric models
Effect.
By formula（1）, can push type（3）.
By formula（3）Can be seen that, when distance parameter t (x) very little, schemed by input based on demisting result J (x) of Atmospheric models
The impact of picture I (x) is very big.In actual scene, distance parameter t (x) very little of sky portion, the demisting result pair of Atmospheric models
Discontinuous and noise in input picture I (x) is very sensitive, and these pixels are located at large stretch of sky areas, demisting knot
Discontinuous and noise can be clearly in fruit.But because sky portion is in itself canescence, mist is also canescence, so to removing
The impact of mist result seems unobvious.Additionally, method of the prior art using threshold value is taken, retains a small amount of mist, see result
Get up more natural.Such as formula（4）It is shown.Threshold value is taken as t0.
And in the scene of sand and dust, or in the scene of colored mist, using the demisting algorithm based on Atmospheric models, and carry out
After the white balance of prior art, sky portion can produce obvious factitious result.For example, colour little in sky is made an uproar
Sound becomes obvious, goes the sand and dust day after tomorrow of the large stretch of yellow of residual in the air, and the aerial bulk in day that caused due to threshold value t0 is not connected
Continue.
The content of the invention
It is an object of the invention to provide a kind of image demister and a kind of image haze removal method, described image demisting dress
Put and can guarantee in different actual scenes good except fog effect with described image defogging method.
It is an object of the invention to provide a kind of image demister, it is characterised in that possess：Input block, input is former
Beginning image；First thick distance parameter asks for unit, calculates the thick distance parameter of each pixel in the original image；Ask at edge
Unit is taken, the marginal information of the original image is calculated；Sky division unit, based on the marginal information and the thick distance ginseng
Number, divides the sky areas in the original image；Second thick distance parameter asks for unit, defeated according to the sky division unit
The sky division result for going out, corrects the thick distance parameter；Thick distance parameter becomes more meticulous unit, to the thick distance parameter being corrected
Become more meticulous；Processing unit, using being entered by the thick distance parameter thick distance parameter that unit become more meticulous that becomes more meticulous
The process of row demisting and white balance are processed.
Image demister of the invention, can make full use of the characteristic information of image, for sand and dust scene or
The scene of person's colour mist, obtains preferably except fog effect, prevents image fault.
In addition, in the image demister of the present invention, the sky division unit judges current according to the marginal information
Whether pixel is located at edge, if located in edge, then judge current pixel and its top to non-sky portion multiple pixels as
Non- sky portion.According to such structure, marginal information can be better profited from, more accurately divide sky areas, so as to
Obtain preferably removing fog effect.
In addition, in the image demister of the present invention, the sky division unit reads the thick distance parameter of current pixel,
In the case where current pixel is not located at edge and its thick distance parameter is more than setting, then judge that current pixel and its top are arrived
Multiple pixels of non-sky portion are non-sky portion.According to such structure, can be effectively combined marginal information and it is thick away from
From parameter, sky areas are more accurately divided, so as to obtain preferably removing fog effect.
In addition, in the image demister of the present invention, the sky division unit reads the thick distance parameter of current pixel,
Edge is not located in current pixel and its thick distance parameter is less than above setting and current pixel and does not exist in preset range
In the case of edge, the pixel is judged to into sky portion.According to such structure, marginal information and thick can be effectively combined
Distance parameter, more accurately divides sky areas, so as to obtain preferably removing fog effect.
In addition, in the image demister of the present invention, the sky division unit reads the thick distance parameter of current pixel,
Edge is not located in current pixel and its thick distance parameter is less than preset range internal memory above the setting and current pixel
In the case of edge, the top edges in the marginal information of current pixel are found, and compare the thick distance of current pixel
The thick distance parameter of parameter and the top edges, if the thick distance parameter of current pixel is close or larger than its top edges
Thick distance parameter, then judge that current pixel is non-sky portion, otherwise, then judges the pixel as sky portion.According to such
Structure, can be effectively combined marginal information and thick distance parameter, more accurately divide sky areas, so as to obtain preferably
Except fog effect.
In addition, in the image demister of the present invention, if deducting current pixel from the thick distance parameter of top edges
The calculated value of thick distance parameter is less than 0.02, then sky division unit is judged to that the thick distance parameter of current pixel is close or larger than
The thick distance parameter of its top edges.According to such structure, marginal information and thick distance parameter can be effectively combined, be reduced
The error that sky is divided, so as to obtain preferably removing fog effect.
In addition, in the image demister of the present invention, when the described second thick distance parameter asks for unit and receive to be determined
For the pixel of sky portion, then the second thick distance parameter asks for unit and the thick distance parameter of the pixel is modified to into predetermined value.
According to such structure, the sky portion affected by mist can be reduced into original pixel, difference can be flexibly be adapted to
Scene, so as to guarantee except being adaptive selected different solutions in the case of fog effect.
In addition, in the image demister of the present invention, the predetermined value is 0.98.According to such structure, can not lose
Sky portion very is reduced, so as to obtain preferably removing fog effect.
In addition, in the image demister of the present invention, the first thick distance parameter ask for unit and it is described second it is thick away from
From parameter unit is asked for for same unit.According to such structure, the quantity of Component units can be reduced, simplification figure is filled as demisting
The structure put.
Another object of the present invention is to a kind of image haze removal method is provided, including：Input original image；Calculate the original
The thick distance parameter of each pixel in beginning image；Calculate the marginal information of the original image；Based on the marginal information and
The thick distance parameter, divides the sky areas in the original image；According to sky division result, the thick distance ginseng is corrected
Number；Thick distance parameter to being corrected becomes more meticulous；Using the thick distance parameter for being become more meticulous carry out demisting process and
White balance process.Image haze removal method of the invention, can make full use of the characteristic information of image, for the scene of sand and dust
Or the scene of colored mist, obtain preferably except fog effect, prevent image fault.
In addition, in the image haze removal method of the present invention, judge whether current pixel is located at edge according to the marginal information,
If located in edge, then judge current pixel and its top to multiple pixels of non-sky portion as non-sky portion.Thus, energy
Marginal information is enough better profited from, sky areas are more accurately divided, so as to obtain preferably removing fog effect.
In addition, in the image haze removal method of the present invention, reading the thick distance parameter of current pixel, it is not located in current pixel
Edge and its thick distance parameter are more than in the case of setting, then judging current pixel and its top to the multiple of non-sky portion
Pixel is non-sky portion.Thereby, it is possible to be effectively combined marginal information and thick distance parameter, day dead zone is more accurately divided
Domain, so as to obtain preferably removing fog effect.
In addition, in the image haze removal method of the present invention, reading the thick distance parameter of current pixel, it is not located in current pixel
In the case of there is no edge less than setting and in the preset range of current pixel top in edge and its thick distance parameter, will
The pixel is judged to sky portion.Thereby, it is possible to be effectively combined marginal information and thick distance parameter, day is more accurately divided
Dummy section, so as to obtain preferably removing fog effect.
In addition, in the image haze removal method of the present invention, reading the thick distance parameter of current pixel, it is not located in current pixel
In the case of there is edge less than the setting and in the preset range of current pixel top in edge and its thick distance parameter,
The top edges in the marginal information of current pixel are found, and compares the thick distance parameter of current pixel and the top side
The thick distance parameter of edge, if the thick distance parameter of current pixel is sentenced close or larger than the thick distance parameter of its top edges
Settled preceding pixel is non-sky portion, otherwise, then judges the pixel as sky portion.Thereby, it is possible to be effectively combined edge letter
Breath and thick distance parameter, more accurately divide sky areas, so as to obtain preferably removing fog effect.
In addition, in the image haze removal method of the present invention, if deducting current pixel from the thick distance parameter of top edges
The calculated value of thick distance parameter is less than 0.02, then be judged to the thick distance parameter of current pixel close or larger than its top edges
Thick distance parameter.Thereby, it is possible to be effectively combined marginal information and thick distance parameter, the error that sky is divided is reduced, so as to
To preferably except fog effect.
In addition, in the image haze removal method of the present invention, when the pixel that is judged as sky portion is received, then by the pixel
Thick distance parameter be modified to predetermined value.Thereby, it is possible to the sky portion affected by mist is reduced into into original pixel, can be with spirit
It is applied to different scenes livingly, so as to guarantee to be adaptive selected different solutions in the case of fog effect.
In addition, in the image haze removal method of the present invention, the predetermined value is 0.98.Thereby, it is possible to reduce day without distortion
Empty part, so as to obtain preferably removing fog effect.
Image haze removal method of the invention or image demister have advantages below：
The characteristic information of image is made full use of, for the scene or the scene of colored mist of sand and dust, more preferable demisting is obtained
Effect, prevents image fault；
Can be easily incorporated in other defogging methods according to image haze removal method of the present invention；
The method according to the invention or device can be realized easily by software or hardware, it is only necessary to for existing soft
Less change is done on the basis of part or hardware；
The method according to the invention can flexibly be adapted to different scenes, so as in the case where guaranteeing except fog effect
It is adaptive selected different solutions.
Description of the drawings
Fig. 1 illustrates the schematic diagram of the Atmospheric models of prior art.
Fig. 2 illustrates the structured flowchart of the image processing system of the image demister 200 including the present invention.
Fig. 3 A illustrate the flow chart of each step of the image haze removal method of an embodiment of the invention.
Fig. 3 B illustrate the example of image resulting in each step in the image haze removal method of Fig. 3 A.
Fig. 4 illustrates the flow chart of method and step S303 in the image haze removal method of Fig. 3 A.
Fig. 5 A, 5B illustrate the thick distance parameter figure and marginal information of the example of three pixels in the image haze removal method of Fig. 4
Figure.
Fig. 6 illustrates the flow chart of method and step S304 in the image haze removal method of Fig. 3 A.
Specific embodiment
With reference to the accompanying drawings and examples the present invention will be described in more detail.
Describe the preferred of image demister involved in the present invention and image haze removal method in detail referring to the drawings
Embodiment.Additionally, in the description of the drawings, same symbol is accompanied by same or considerable part, the repetitive description thereof will be omitted.
Following examples are related to the single-frame images demister based on Atmospheric models and method.But, the present invention's also may be used
With suitable for the image demister and method based on other models.Additionally, the image demister and method of the present invention include
Demisting, except sand and dust, except the extensive image demister such as Atmospheric particulates and method.
Fig. 2 illustrates the structured flowchart of the image processing system of the image demister 200 including the present invention.According to the present invention
Image processing system include camera head（That is, input block）100th, image demister 200, shared memory 300 and defeated
Go out unit 400.The camera head 100 is used to absorb image and described image is transferred to into image demister 200.It is described
Image demister 200 is used for the image to being provided by camera head 100 carries out sharpening process（Also referred to as demisting is processed）.Institute
Shared memory 300 is stated for storing various data.The output unit 400 is used to export（And/or show）Jing image demistings
The image of the sharpening of device 200 process.
Image demister 200 including pretreatment unit 10, edge ask for unit 20, thick distance parameter ask for unit 30,
Sky division unit 40, thick distance parameter become more meticulous unit 50, image demisting unit 60 and control unit 70.
The pretreatment unit 10 is used to obtain the current frame image that analysis is provided by camera head 100, calculates present frame
The day null point parameter of image.Current frame image is transferred to edge and asks for unit 20 by the pretreatment unit 10, present frame figure
Picture and the day null point parameter for calculating are transferred to the thick distance parameter and ask for unit 30.
Unit 20 is asked for for asking for the marginal information of current frame image in the edge.Asking for image edge information can make
Use existing Edge-Detection Algorithm, such as canny algorithms.
The thick distance parameter asks for the current frame image and day null point parameter that unit 30 reads the output of pretreatment unit 10,
Calculate thick distance parameter.
The sky division unit 40 reads edge and asks for the marginal information of the output of unit 20, and reads thick distance parameter and ask
The thick distance parameter of the output of unit 30 is taken, the division result of sky portion is realized in calculating.
The thick distance parameter asks for the sky division result that unit 30 reads the output of sky division unit 40, arranges new
Thick distance parameter.
The thick distance parameter unit 50 that becomes more meticulous reads the new thick distance ginseng that thick distance parameter asks for the output of unit 30
Number, is become more meticulous using prior art, such as scratch nomography.
Processing unit 60 to current frame image by the thick distance parameter smart distance parameter that unit 50 obtains that becomes more meticulous using being entered
The post processing such as the process of row demisting and white balance.Here, for example institute can be carried out by the known defogging method based on Atmospheric models
State demisting process.Preferably, the image for being provided by camera head 100 passes through between camera head 100 and processing unit 60（Not
Illustrate）Connection is transferred to processing unit 60.Alternatively, processing unit 60 can also be from memory --- such as shared memory
Current frame image is read in 300.
Control unit 70 is used to controlling or configuring the unit or module in image demister 200.
Fig. 3 A illustrate the signal of the image haze removal method of the image demister 200 of an embodiment of the invention
Figure.Fig. 3 B illustrate the example of image resulting in each step in the image haze removal method of Fig. 3 A.
Original image is as shown in the first row in the table of Fig. 3 B.
In method and step S301, using prior art thick distance parameter is calculated.In the table of resulting image such as Fig. 3 B
The second row shown in.
In method and step S302, using prior art rim detection is carried out.In the table of resulting image such as Fig. 3 B
Shown in the third line.
In ensuing method and step S303, according to thick distance parameter and marginal information, sky division is carried out.It is resulting
Image as shown in the fourth line in the table of Fig. 3 B.It can be seen that sky portion is clearly marked off coming.
In ensuing method and step S304, according to thick distance parameter and sky division result, new thick distance is calculated
Parameter.Resulting image is as shown in the fifth line in the table of Fig. 3 B.
In method and step S305, new thick distance parameter is become more meticulous using prior art.Resulting image
As shown in the 6th row in the table of Fig. 3 B.
In method and step S306, using prior art demisting and white balance are realized.The table of resulting image such as Fig. 3 B
In last column shown in.Output image is compared with input picture, it can be found that the aerial sand and dust in day are significantly removed,
In addition, the sky portion below building is also specifically identified, and complete demisting process.
Fig. 4 illustrates the flow chart of method and step S303 in the image haze removal method of Fig. 3 A.Fig. 5 A, 5B illustrate the image of Fig. 4
The thick distance parameter figure and marginal information figure of the example of three pixels in defogging method.
In method and step S400, judge whether the last pixel for reaching image.
If judging to have reached last pixel in method and step S400, followed by method and step S410, output
Sky division result.
If judging not reaching last pixel in method and step S400, followed by method and step S402, that is, cut
Change to next processes pixel.
In method and step S403, judge whether current pixel is located at the marginal position of image（Marginal information is walked by method
Rapid S302 is obtained）.
If judging that current pixel is located at edge in method and step S403, followed by method and step S408.
If judging that current pixel is not located at edge in method and step S403, followed by method and step S404.
In ensuing method and step S404, the thick distance parameter of current pixel is judged（Thick parameter）Whether less than default
Value（The scope of such as preset value is 0.6~0.7, preferably 0.65）.
If the thick distance parameter of current pixel is judged in method and step S404 more than preset value, followed by step
Rapid S408.
If the thick distance parameter of current pixel is judged in method and step S404 less than preset value, followed by side
Method step S405.For example, in Fig. 5 A, 5B three pixels 1.~be 3. satisfied by this condition, then for these three pixels are walked
The operation of rapid S405.
Preset range above current pixel is judged in method and step S405（Such as 3 pixel coverages）It is interior with the presence or absence of side
Edge.
If judging there is no edge above current pixel in preset range in method and step S405, current pixel is sentenced
Break as sky portion, followed by method and step S409.For example, the pixel in Fig. 5 A, 5B is 1..
If judging there is edge above current pixel in preset range in method and step S405, followed by side
Method step S406.For example, the pixel in Fig. 5 A, 5B is 2., 3..
Judge in method and step S406 current pixel thick distance parameter value whether be more than or close its top edges at
The thick distance parameter value of pixel.
If judging that the thick distance parameter value of current pixel is little square thereon to a certain extent in method and step S406
The thick distance parameter value of edge pixel（Such as at the top edges thick distance ginseng of the thick distance parameter value-current pixel of pixel
Numerical value>0.02）, then current pixel be judged as sky portion, followed by method and step S409.For example, the picture in Fig. 5 A, 5B
Element is 3..
If judge in method and step S406 current pixel thick distance parameter value be more than or close its top edges at
The thick distance parameter value of pixel（Such as at the top edges thick distance parameter value of the thick distance parameter value-current pixel of pixel<
0.02）, then followed by method and step S407, that is, current pixel is marked to be non-sky portion.For example, the picture in Fig. 5 A, 5B
Element is 2..
In following step S408, current pixel and its top is marked to be non-until multiple pixels of non-sky portion
Sky portion.
Next method and step S401 is returned to, judges whether to reach last pixel.If it is, output sky divides knot
Really, otherwise proceed as described above.
According to above-mentioned steps, sky portion can be accurately divided, and no matter original image is sand and dust or colored mist（Example
Such as the blue mist on seashore）In the environment of, can effectively divide sky portion.Also, it is pre- in by changing step S404
If value and the preset range in step S405, can adapt to a variety of images.
Fig. 6 illustrates the flow chart of method and step S304 in the image haze removal method of Fig. 3.
In method and step S501, the thick distance parameter image of read method step S301 output.
In method and step S502, the sky of read method step S303 output divides image.
In method and step S503, the sky obtained in read method step S502 divides current pixel position in image
Value.
In ensuing method and step S504, whether the value of determination methods step S503 output is labeled as sky portion.
If judging that the value is labeled as sky portion in method and step S504, followed by method and step S505.
If judging that the value is labeled as non-sky portion in method and step S504, followed by method and step
S506。
In method and step S505, respective pixel position in the thick distance parameter image read in amending method step S501
Thick distance parameter value be predetermined value.0.98 can be for example set to.Followed by method and step S506.
In method and step S506, judge whether to reach last pixel.
If judging to have reached last pixel in method and step S506, followed by method and step S508, that is, protect
Deposit new thick distance parameter image.
If judging not reaching last pixel in method and step S506, followed by method and step S507, i.e., after
It is continuous to process next pixel.
Method according to Fig. 6, can be modified the thick distance parameter of the pixel of sky portion, even if so
In the case that sky portion is shrouded by haze, it is also possible to reduce the color of sky script, make image apparent.
The image demister or image processing system of the present invention is particularly well-suited to field of video monitoring, while can also use
In any equipment related to image, video, the camera head as, camera etc..
Although being illustrated to the present invention above in association with drawings and Examples, it will be appreciated that described above
The invention is not limited in any way.For example, same thick distance parameter has been recorded in above-mentioned embodiment unit 30 has been asked for and read
The current frame image and day null point parameter of the output of pretreatment unit 10 are taken, thick distance parameter is calculated；Also, read sky and divide single
The sky division result of the output of unit 40, arranges new thick distance parameter.But it is also possible to ask for unit by two thick distance parameters
The two steps are carried out respectively.Those skilled in the art without departing from the true spirit and scope of the present invention can be with root
The present invention is deformed and is changed according to needing, these deformations and change are within the scope of the present invention.
Claims (9)
1. a kind of image demister, it is characterised in that
Possess：
Input block, is input into original image；
First thick distance parameter asks for unit, calculates the thick distance parameter of each pixel in the original image；
Unit is asked at edge, calculates the marginal information of the original image；
Sky division unit, based on the marginal information and the thick distance parameter, divides the day dead zone in the original image
Domain；
Second thick distance parameter asks for unit, according to the sky division result of sky division unit output, corrects described thick
Distance parameter；
Thick distance parameter becomes more meticulous unit, and the thick distance parameter to being corrected becomes more meticulous；
Processing unit, using by the thick distance parameter thick distance parameter that unit become more meticulous that becomes more meticulous demisting is carried out
Reason and white balance are processed,
The sky division unit judges that whether current pixel is located at edge according to the marginal information, if located in edge, then
Judge current pixel and its top to multiple pixels of non-sky portion as non-sky portion,
The sky division unit reads the thick distance parameter of current pixel, and in current pixel edge and its thick distance ginseng are not located at
Number then judges current pixel and its top to multiple pixels of non-sky portion as non-sky portion more than in the case of setting
Point,
The sky division unit reads the thick distance parameter of current pixel, and in current pixel edge and its thick distance ginseng are not located at
In the case that number does not have edge less than setting and in the preset range of current pixel top, the current pixel is judged to
Sky portion,
The sky division unit reads the thick distance parameter of current pixel, and in current pixel edge and its thick distance ginseng are not located at
In the case that number has edge less than the setting and in the preset range of current pixel top, the institute of current pixel is found
The top edges in marginal information are stated, and compare the thick distance parameter of current pixel and the thick distance parameter of the top edges,
If the thick distance parameter of current pixel is slightly smaller or larger than the thick distance parameter of its top edges, current pixel right and wrong are judged
Sky portion, otherwise, then judges the current pixel as sky portion,
The thick distance parameter is tried to achieve according to Atmospheric models formula,
I (X)=J (X) t (X)+A (1-t (X))
Wherein, I (X) represents the atomization image that camera head photographs or the atomization image that eye-observation is arrived, and X=(x, y) is
Image pixel coordinates, J (X) is object reflected light image, represents the image without mist, A for day null point in image rgb value, t
(X) it is one and is more than 0 and the scalar data less than 1, referred to as distance parameter.
2. image demister as claimed in claim 1, it is characterised in that
If the calculated value of the thick distance parameter of current pixel is deducted from the thick distance parameter of top edges less than 0.02, sky
Division unit is judged to that the thick distance parameter of current pixel is slightly smaller or larger than the thick distance parameter of its top edges.
3. image demister as claimed in claim 1, it is characterised in that
Unit is asked for when the described second thick distance parameter and receive the pixel that is judged as sky portion, then the second thick distance ginseng
Number asks for unit and the thick distance parameter of the pixel for being judged as sky portion is modified to into predetermined value.
4. image demister as claimed in claim 3, it is characterised in that
The predetermined value is 0.98.
5. the image demister as any one of Claims 1 to 4, it is characterised in that
The first thick distance parameter asks for unit and the second thick distance parameter asks for unit for same unit.
6. a kind of image haze removal method, it is characterised in that
Including：
Input original image；
Calculate the thick distance parameter of each pixel in the original image；
Calculate the marginal information of the original image；
Based on the marginal information and the thick distance parameter, the sky areas in the original image are divided；
According to sky division result, the thick distance parameter is corrected；
Thick distance parameter to being corrected becomes more meticulous；
Demisting is carried out using the thick distance parameter for being become more meticulous to process and white balance process,
Judge that whether current pixel is located at edge according to the marginal information, if located in edge, then judge current pixel and its
Top to multiple pixels of non-sky portion are non-sky portion,
The thick distance parameter of current pixel is read, edge is not located in current pixel and its thick distance parameter is more than the feelings of setting
Under condition, then judge current pixel and its top to multiple pixels of non-sky portion as non-sky portion,
Read the thick distance parameter of current pixel, be not located at edge and its thick distance parameter less than setting in current pixel and
And current pixel top preset range in there is no edge in the case of, the current pixel is judged to into sky portion,
The thick distance parameter of current pixel is read, edge is not located in current pixel and its thick distance parameter is less than the regulation
In the case of there is edge in value and current pixel top preset range, in finding the marginal information of current pixel
Top edges, and compare the thick distance parameter of current pixel and the thick distance parameter of the top edges, if current pixel
Thick distance parameter is slightly smaller or larger than the thick distance parameter of its top edges, then judge that current pixel is non-sky portion, otherwise,
The current pixel is then judged as sky portion,
The thick distance parameter is tried to achieve according to Atmospheric models formula,
I (X)=J (X) t (X)+A (1-t (X))
Wherein, I (X) represents the atomization image that camera head photographs or the atomization image that eye-observation is arrived, and X=(x, y) is
Image pixel coordinates, J (X) is object reflected light image, represents the image without mist, A for day null point in image rgb value, t
(X) it is one and is more than 0 and the scalar data less than 1, referred to as distance parameter.
7. image haze removal method as claimed in claim 6, it is characterised in that
If the calculated value of the thick distance parameter of current pixel is deducted from the thick distance parameter of top edges less than 0.02, judge
Thick distance parameter for current pixel is slightly smaller or larger than the thick distance parameter of its top edges.
8. image haze removal method as claimed in claim 6, it is characterised in that
When the pixel that is judged as sky portion is received, then the thick distance parameter for being judged as the pixel of sky portion is repaiied
Just it is being predetermined value.
9. image haze removal method as claimed in claim 8, it is characterised in that
The predetermined value is 0.98.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310106782.2A CN104077745B (en) | 2013-03-29 | 2013-03-29 | Image demisting device and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310106782.2A CN104077745B (en) | 2013-03-29 | 2013-03-29 | Image demisting device and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104077745A CN104077745A (en) | 2014-10-01 |
| CN104077745B true CN104077745B (en) | 2017-05-17 |
Family
ID=51598988
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310106782.2A Expired - Fee Related CN104077745B (en) | 2013-03-29 | 2013-03-29 | Image demisting device and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104077745B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106934812A (en) * | 2015-12-25 | 2017-07-07 | 北京展讯高科通信技术有限公司 | Image-signal processor and its image-signal processing method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102930514A (en) * | 2012-09-27 | 2013-02-13 | 西安电子科技大学 | Rapid image defogging method based on atmospheric physical scattering model |
| EP2568438A2 (en) * | 2011-09-08 | 2013-03-13 | Fujitsu Limited | Image defogging method and system |
-
2013
- 2013-03-29 CN CN201310106782.2A patent/CN104077745B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2568438A2 (en) * | 2011-09-08 | 2013-03-13 | Fujitsu Limited | Image defogging method and system |
| CN102930514A (en) * | 2012-09-27 | 2013-02-13 | 西安电子科技大学 | Rapid image defogging method based on atmospheric physical scattering model |
Non-Patent Citations (2)
| Title |
|---|
| Haze filtering with aerial perspective;Renjie Gao et al;《19th IEEE International Conference on Image Processing》;20120930;第989页右栏第2.1节、第2.2节，第991页左栏3.2节、右栏第4.2节，图4 * |
| 一种改进的暗原色单幅图像去雾方法;陈瑶 等;《四川理工学院学报(自然科学版)》;20121031;第25卷(第5期);62-64 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104077745A (en) | 2014-10-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106296612B (en) | A kind of stagewise monitor video sharpening system and method for image quality evaluation and weather conditions guidance | |
| US11087151B2 (en) | Automobile head-up display system and obstacle prompting method thereof | |
| CN105354806B (en) | Rapid defogging method and system based on dark | |
| CN103955905B (en) | Based on the single image to the fog method that fast wavelet transform and weighted image merge | |
| CN102750674B (en) | Video image defogging method based on self-adapting allowance | |
| CN101783012B (en) | Automatic image defogging method based on dark primary colour | |
| TWI501194B (en) | Method and image prcoessing apparatus for image visibility restoration using fisher'a linear discriminant based dual dark channel prior | |
| US9384532B2 (en) | Apparatus for improving fogged image using user-controllable root operator | |
| CN102831591B (en) | Gaussian filter-based real-time defogging method for single image | |
| CN103186887B (en) | Image demister and image haze removal method | |
| CN102289791B (en) | Method for quickly demisting single image | |
| CN108734670A (en) | The restoration algorithm of single width night weak illumination haze image | |
| CN105913390B (en) | A kind of image defogging method and system | |
| CN104751421B (en) | A kind of image defogging method realized on FPGA | |
| CN105701783B (en) | A kind of single image to the fog method and device based on environment light model | |
| CN104867121B (en) | Image Quick demisting method based on dark primary priori and Retinex theories | |
| CN103034983B (en) | A kind of defogging method capable based on anisotropic filtering | |
| CN107424133A (en) | Image defogging method, device, computer can storage medium and mobile terminals | |
| CN106355560B (en) | The extracting method and system of air light value in a kind of haze image | |
| CN107451966A (en) | A kind of real-time video defogging method realized using gray-scale map guiding filtering | |
| CN111179189A (en) | Image processing method and device based on generation countermeasure network GAN, electronic equipment and storage medium | |
| CN106296618A (en) | A kind of color image defogging method based on Gaussian function weighted histogram regulation | |
| CN103226809B (en) | Image demister and image haze removal method | |
| CN107277299A (en) | Image processing method, device, mobile terminal and computer-readable recording medium | |
| CN111563852A (en) | Dark channel prior defogging method based on low-complexity MF |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20171206 Address after: Section Information Center No. 2 Beijing Road, Haidian District Academy of Sciences C North Building 301 Patentee after: Hitachi (China) Research & Development Corporation Address before: Tokyo, Japan, Japan Patentee before: Hitachi Ltd. |
|
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170517 Termination date: 20200329 |