CN111539891A - Wave band self-adaptive demisting optimization processing method for single remote sensing image - Google Patents
Wave band self-adaptive demisting optimization processing method for single remote sensing image Download PDFInfo
- Publication number
- CN111539891A CN111539891A CN202010345923.6A CN202010345923A CN111539891A CN 111539891 A CN111539891 A CN 111539891A CN 202010345923 A CN202010345923 A CN 202010345923A CN 111539891 A CN111539891 A CN 111539891A
- Authority
- CN
- China
- Prior art keywords
- image
- remote sensing
- fog
- defogging
- adaptive
- 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
- 238000005457 optimization Methods 0.000 title claims abstract description 31
- 238000003672 processing method Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 127
- 238000012545 processing Methods 0.000 claims abstract description 64
- 230000003595 spectral effect Effects 0.000 claims abstract description 32
- 239000003595 mist Substances 0.000 claims abstract description 26
- 230000006872 improvement Effects 0.000 claims abstract description 3
- 238000002834 transmittance Methods 0.000 claims description 53
- 238000001914 filtration Methods 0.000 claims description 27
- 230000008569 process Effects 0.000 claims description 19
- 230000003044 adaptive effect Effects 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 9
- 239000003086 colorant Substances 0.000 claims description 5
- 238000009826 distribution Methods 0.000 claims description 5
- 230000005540 biological transmission Effects 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000003384 imaging method Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 19
- 230000002159 abnormal effect Effects 0.000 description 20
- 238000010586 diagram Methods 0.000 description 8
- 230000000007 visual effect Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 238000012886 linear function Methods 0.000 description 4
- 238000009499 grossing Methods 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 239000002689 soil Substances 0.000 description 2
- 238000005054 agglomeration Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/73—Deblurring; Sharpening
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/10—Image enhancement or restoration using non-spatial domain filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention provides a wave band self-adaptive demisting optimization processing method of a single remote sensing image, which is based on an initial dark primary color prior demisting method of the remote sensing image, and firstly, in order to solve the problem that high-brightness ground objects in the image are abnormally dark after demisting by using the initial dark primary color prior method, firstly, the high-brightness ground objects in the image are extracted by using spectral information, then, the extracted high-brightness ground objects are independently subjected to self-adaptive processing, and the spectral information of the high-brightness ground objects is reserved; and secondly, the processing of different degrees is carried out on each wave band in the remote sensing image, so that the problem that the image is abnormally blue after defogging is effectively avoided. Through the optimization and improvement of the two aspects, cloud and mist pollution can be effectively removed, the spectral information of ground objects can be effectively maintained, and a very satisfactory defogging effect is obtained in general.
Description
Technical Field
The invention relates to a demisting processing method for remote sensing images, in particular to a waveband self-adaptive demisting optimization processing method for a single remote sensing image, and belongs to the technical field of remote sensing image processing.
Background
With the rapid development of remote sensing technology and computer technology, the application of remote sensing images enters a new stage, and the remote sensing images play a wide and important role in numerous fields of current social life and production. However, in practical applications, the remote sensing image is usually affected by climate and environment, especially in areas covered by cloud layers or haze, and the real and accurate reflection of the remote sensing image on the earth surface information is greatly affected by the cloud and haze pollution, which not only visually affects and interferes with manual interpretation, but also has significant negative effects on the precision of various subsequent processes, such as extraction and classification of earth features. In the remote sensing image, due to the area covered by the thick cloud, the ground feature information is almost completely lost, and the thick cloud can be removed only by referring to other auxiliary images or data. The area covered by the haze and the thin cloud not only contains the information of the haze and the thin cloud, but also contains the information of the ground objects, so that the influence of the haze and the thin cloud can be removed through some technical methods, and the real information of the ground objects can be recovered.
The interference of haze and thin cloud in the atmosphere on the remote sensing image is mainly expressed in the following two aspects: on one hand, haze and thin cloud often present local high-brightness characteristics in the remote sensing image, so that partial ground object information can be shielded, the gray scale range of pixels in a polluted area is reduced, and the contrast of the remote sensing image is reduced; on the other hand, due to the interference of haze and thin clouds, the spectral information of the covered ground objects is distorted, the colors of the ground objects are greatly deviated, a satisfactory visual effect is difficult to achieve, and a plurality of problems are brought to the processing of remote sensing images.
After the haze and the thin cloud of the remote sensing image are corrected, the optimized image also meets the following two conditions: firstly, the improved remote sensing image has higher contrast than the initial image, and can embody more ground feature detail information in the image; secondly, the improved remote sensing image can recover the real color information of the polluted ground object in the initial image, so that the improved remote sensing image has more accurate spectral information and more natural visual effect.
Through the two aspects of processing, the quality of the image is improved, the usability of the remote sensing image data is improved, and good bedding is made for visual interpretation and various subsequent processing. Therefore, the technology for removing the haze and the thin cloud in the remote sensing image is researched and developed, the interference of the haze and the thin cloud to the remote sensing image is restrained and eliminated, the usability of the remote sensing data is improved, and the method has important practical significance and application value.
In the prior art, from the nineties, attention is gradually paid to the problem of removing haze and thin clouds in remote sensing images, and a lot of researches on defogging methods in the remote sensing images have been conducted so far. Generally, according to different processing ideas, the defogging methods in the remote sensing images in the prior art can be roughly classified into the following four types:
the demisting method based on characteristic extraction comprises the following steps: the haze and the thin cloud have obvious difference with the ground objects in the spectral domain and the frequency domain, so that the distribution information of the cloud and the mist can be extracted by constructing specific conditions to highlight the difference between the cloud and the ground objects, and then the region polluted by the cloud and the mist is enhanced in a targeted manner to remove the cloud and the mist in the image, for example: homomorphic filtering method, tassel-cap conversion method, wave band ratio algorithm, HIS color space conversion and other defogging methods.
Secondly, a cloud and mist removing method based on multi-source multi-temporal information comprises the following steps: the method mainly comprises the steps of replacing or fusing local areas polluted by cloud and mist by utilizing multi-source or multi-temporal images covering the same area and local areas of pure images which are not polluted by the cloud and mist to achieve the aim of demisting, wherein the methods can be mainly divided into a multi-temporal information demisting method and a multi-source information demisting method.
Thirdly, a band fusion demisting method: cloud and fog in the remote sensing image generally have obvious influence only on three visible wave bands of RGB (red, green and blue), the influence on wave bands with near infrared and longer wavelength is not obvious, and the defogging effect is realized based on wave band fusion by combining spectral information of three wave bands of red, green and blue and gradient information of a long wave band. At present, a defogging method based on gradient fusion is a method which has a better band fusion effect.
Fourthly, demisting method based on prior hypothesis: firstly, summarizing and inducing certain commonalities of the cloud-fog-free images, then deducing prior conditions and assumptions according to the commonalities, and finally restoring the cloud-fog-polluted images into the fog-free images directionally by utilizing the prior conditions and assumptions. Representative methods of this type are: DOS dark target method, HOT optimum cloud and fog detection transformation method.
However, in the defogging methods in the prior art, because the characteristic that fog in the real remote sensing image is unevenly distributed is ignored, and the defogging treatment is performed on the whole image in an integral unified treatment mode, the cloud and fog in partial areas of the defogged image cannot be completely removed, and the problem of spectral distortion is caused by over-treatment in other areas, so that the cloud and fog pollution unevenly distributed in the remote sensing image cannot be removed. Some prior art methods have a major limitation, and when the scene in the remote sensing image is very close to the atmospheric light, the dark channel is not a priori valid because the dark primary value of the object with high brightness in the image cannot be close to 0, and if the initial dark primary method is used, the problem of over-processing occurs, so some prior art methods are not suitable for these scenes. The prior art also finds the remote sensing image processing by using a dark channel prior defogging method, but some outstanding problems still exist, and the most core problems comprise:
firstly, the problem of abnormal and dark high-brightness ground features. After the remote sensing image is demisted by using the initial dark channel prior, cloud and fog pollution in the image can be basically removed, but the demisting of high-brightness ground objects such as bare soil in the image has the obvious problem of abnormal yellow and dark. The reason for this problem is that the features in the scene are themselves high in brightness, and the dark channel prior theory is not applicable to these features when the spectral information is close to atmospheric light. Therefore, if these areas are defogged by the original dark process, the problem of over-treatment occurs.
Secondly, the demisted image is abnormally blue. The common natural image only comprises red, green and blue visible light wave bands generally, while the wave band information of part of remote sensing images is rich, and the common natural image also comprises near infrared wave bands, middle infrared wave bands and far infrared wave bands besides the red, green and blue visible light wave bands. And as for the near infrared band and the intermediate infrared band and the thermal infrared band with the wavelength larger than the near infrared wavelength, the influence of fog is hardly caused, so that when the multi-band remote sensing image is demisted, demisting treatment is only carried out on the three visible light bands of RGB. The image subjected to defogging treatment by using the initial dark channel prior method has the problem of obvious bluish color, and the reason for generating the abnormal bluish color is that the influence degrees of fog on various wave bands are different for multiband images, and the influence degrees of fog on various wave bands are gradually reduced along with the increase of the wavelength. Therefore, the influence degrees of the fog on the red, green and blue wave bands are different, the influence degree of the fog on the blue wave band is the largest, the influence degree on the green wave band is the next, and the influence degree on the red wave band is the smallest. However, the initial dark channel prior method does not take this point into consideration, and the same transmittance d is used for the three visible light bands of RGB to perform the same defogging treatment, so that cloud and mist pollution on the red band after defogging can be completely removed, and cloud and mist pollution on the green band and the blue band can not be completely removed, so that the image after defogging has an abnormal blue tendency.
The invention aims at the problems of high-brightness ground object in the demisted image being abnormal and dark and the demisted image being abnormal and blue, and performs self-adaptive demisting optimization treatment, applies the initial dark primary color prior demisting method to the remote sensing image demisting, and obtains satisfactory demisting effect.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a wave band self-adaptive demisting optimization processing method of a single remote sensing image, which is based on an initial dark primary color prior demisting method of the remote sensing image, firstly, in order to solve the problem that high-brightness ground objects in the image are abnormally dark after demisting by the initial dark primary color prior method, firstly, the high-brightness ground objects in the image are extracted by utilizing spectral information, then, the extracted high-brightness ground objects are independently subjected to self-adaptive processing, and the spectral information of the high-brightness ground objects is reserved; and secondly, the method proposes that different degrees of processing are required to be performed on each wave band in the remote sensing image, the defogging processing on the blue wave band is maximum, the defogging processing on the red wave band is performed to a lesser degree next to the green wave band, and the defogging processing is not performed on the near infrared and long wave bands with wavelengths larger than the near infrared wave band, so that the problem that the image is abnormally blue after defogging is effectively avoided. Through the optimization and improvement of the two aspects, cloud and mist pollution can be effectively removed, the spectral information of ground objects can be effectively maintained, and a very satisfactory defogging effect is obtained in general.
In order to achieve the technical effects, the technical scheme adopted by the invention is as follows:
the wave band self-adaptive demisting optimization processing method of a single remote sensing image is improved on the basis of an initial dark channel prior demisting method,
on one hand, a processing method for solving the problem that high-brightness ground objects in an image are abnormally dark after defogging by using an initial dark primary color prior method is provided: adopting high-brightness ground object self-adaptive demisting treatment, firstly extracting high-brightness ground objects in a remote sensing image by using spectral information, and then independently carrying out self-adaptive treatment on the extracted high-brightness ground objects to keep the spectral information of the high-brightness ground objects;
on the other hand, a processing method for solving the problem that the image after defogging processing is abnormally blue by using an initial dark channel prior method is provided: adopting wave band self-adaptive demisting treatment, carrying out different degrees of treatment on each wave band in the remote sensing image, wherein the demisting treatment on a blue wave band is maximum, the demisting treatment on a red wave band is carried out next to a green wave band to a lesser extent, and the demisting treatment is not carried out on a long wave band of which the near infrared and wavelength are greater than the near infrared wave band;
the wave band self-adaptive demisting optimization processing method of the single remote sensing image comprises the following steps:
firstly, obtaining an initial remote sensing image;
secondly, carrying out standardized processing on the remote sensing image;
thirdly, calculating dark primary colors;
fourthly, generating a dark primary color image;
fifthly, calculating atmospheric light;
sixthly, calculating initial color transmittance;
seventhly, performing high-brightness ground object self-adaptive demisting treatment and wave band self-adaptive demisting treatment;
eighthly, optimizing the transmissivity of the guide image filtering drive;
and ninthly, solving the remote sensing image after defogging.
The invention relates to a wave band self-adaptive demisting optimization processing method of a single remote sensing image, in particular to a method for removing cloud and fog pollution which is unevenly distributed in a remote sensing image based on an initial dark primary color prior demisting method;
the content of the initial dark channel prior demisting method is as follows: for most outdoor fog-free remote sensing images, a sky high brightness area is removed, a plurality of pixels always exist in a local area of the image, the intensity values of the pixels on one or a plurality of color channels of RGB (red, green and blue) three color channels are very low and are close to 0, namely, a local area is arbitrarily selected on the outdoor fog-free image, the minimum value on R, G, B color channels in the local area is counted, and the value is close to 0;
randomly selecting an outdoor fog-free image A, and counting dark primary color values A in a local area of the fog-free image AdarkValue of dark primary color AdarkRepresented by formula 1:
in the formula 1, AcOne color channel representing the image, r, g, b are color channels, Ω (x) represents a square local area of the image centered on a pixel located at x;
for most outdoor fog-free images, except for areas with high brightness of the sky, the locally obtained dark primary color intensity value in the image is very low and is close to 0, and formula 2 is described as follows:
Adark→ 0 formula 2
Describing the composition of the fog image by using a structural equation of the fog image, wherein the equation is expressed in the form of formula 3:
b (x) ═ a (x) d (x) + E [1-d (x) ] formula 3
In formula 3, b (x) represents a fog image observed in reality, a (x) is an ideal fog-free scene and is also a result of the desired final restoration, E is an atmospheric light value in the global range and represents an intensity value of a pixel with the maximum fog concentration in the whole image, d (x) represents the transmittance of the image, namely, the light which is not scattered by the atmosphere in the transmission process and finally enters the imaging system accounts for the proportion of the total light;
the structural equation of the fog image shows that a fog remote sensing image is composed of two parts, one part is a real fog-free scene, the other part is interference of atmospheric environment light, and the weights of the two parts are d (x) and 1-d (x) respectively.
The wave band self-adaptive demisting optimization processing method of a single remote sensing image is characterized in that further, the atmospheric light E represents the pixel value of the point with the maximum fog concentration in the whole image, and the pixel value is obtained according to the following two steps:
step 1, taking 15 multiplied by 15 as a calculation step length to obtain a dark primary color image of an initial image, and selecting a front one thousandth brightest pixel in the dark primary color image;
and 2, corresponding one thousandth of brightest pixels to the initial image, selecting the pixels with the maximum brightness values, and taking the RGB values of the pixels with the maximum brightness values as the values of the atmospheric light E.
A wave band self-adaptive demisting optimization processing method for a single remote sensing image further comprises the steps of supposing that atmospheric light E is obtained, B represents a foggy image observed in reality, A is an ideal fogless scene, and AcRepresenting one color channel of an image, r, g, b are color channels, Ω (x) represents a square local area of the image centered on a pixel located at x, and in the local area Ω (x), the transmittance d' (x) is a constant value, and equation 4 can be found by combining equation 1 and equation 3:
a wave band self-adaptive demisting optimization processing method for a single remote sensing image further combines a dark primary color prior theory to obtain a local area dark primary color A in an outdoor fog-free image AdarkIs close to 0, while the atmospheric light value E in the global range is constantly positive, according to the above two conditions, equation 5 can be obtained:
the value of the transmittance d' (x) is obtained by substituting equation 5 into equation 4, as shown in equation 6:
after the normalization process is performed on the initial image,
namely representing the dark primary color value of a local area in the image;
finally, after obtaining the values of the atmospheric light C and the transmittance d' (x), and optimizing the initial transmittance map, the fog-free image a can be recovered by integrating equation 3 and equation 6:
through the steps, the dark channel prior theory is combined with the structural equation of the fog image, and the fog pollution in the remote sensing image is removed.
The invention discloses a wave band self-adaptive demisting optimization processing method of a single remote sensing image, and further provides a high-brightness ground object self-adaptive demisting processing method, which comprises the following steps: firstly, extracting high-brightness ground objects in a remote sensing image according to different spectral characteristics of the ground objects; then, the extracted high-brightness ground object is subjected to self-adaptive defogging treatment, and appropriate defogging intensity is given.
A wave band self-adaptive demisting optimization processing method of a single remote sensing image is characterized in that further, in the extraction of high-brightness ground objects, if an image only has three RGB wave bands, according to the difference of spectral characteristics of each ground object, a discrimination condition F is adopted to extract the high-brightness ground objects in the remote sensing image:
in the formula 8, RGB represents three bands of red, green and blue of the image, respectively;
judging each pixel in the remote sensing image by using the judgment condition of the formula 8, and judging the pixel as a high-brightness ground object if the F value corresponding to the pixel is less than 1; if the F value is more than 1.5, judging the ground object as non-high-brightness ground object; if the F value is between 1 and 1.5, judging that the transition from the high-brightness ground object to the non-high-brightness ground object is carried out;
if the remote sensing image has a plurality of wave bands such as near infrared besides three wave bands of red, green and blue, the high-brightness ground object formula in the image is accurately extracted by using two coefficients of the normalized water index NDWI and the normalized vegetation index NDVI as judgment conditions,
in formulas 9 and 10, Nir represents a near infrared band, R represents a red band, and G represents a green band.
A wave band self-adaptive defogging optimization processing method for a single remote sensing image further comprises the steps of carrying out self-adaptive defogging processing on high-brightness ground objects, extracting the high-brightness ground objects in the remote sensing image, carrying out defogging processing, endowing the high-brightness ground objects with proper defogging intensity, carrying out self-adaptive defogging processing, introducing a parameter h to adjust the defogging processing intensity, wherein the value range of h is between 0 and 1, d' (x) is the corrected transmissivity,
for an image with three RGB bands, the value of h is shown in formula 12:
the transmissivity of the high-brightness ground area is corrected by introducing a parameter h, the processing intensity of the high-brightness ground object in the remote sensing image is adjusted, and the corrected transmissivity image is subjected to smooth filtering processing to obtain a final transmissivity image;
the atmospheric light E and the corrected transmittance d' (x) are solved, and then the image is finally subjected to defogging processing according to equation 11.
The invention provides a wave band self-adaptive demisting optimization processing method of a single remote sensing image, and further, in the wave band self-adaptive demisting processing, demisting processing is only carried out on three visible wave bands of RGB (red, green and blue), and the invention provides a model: after the transmissivity is calculated through the initial dark primary color, the transmissivity aiming at each wave band of RGB is calculated, and self-adaptive demisting treatment with different degrees is carried out on the three wave bands of RGB;
meanwhile, the distribution concentration of the cloud mist is assumed, and the following two conditions are met on three RGB wave bands:
firstly, to the cloud and fog of the same concentration, to red green blue three wave band's influence degree increase progressively, the transmissivity of the three wave bands of RGB ranks as: the red band has the highest transmittance, the green band is the second order, and the blue band is the lowest; therefore, in the foggy area, the processing intensity of the RGB three wave bands is increased progressively;
secondly, the more dense the concentration of the cloud mist, the larger the influence difference of the cloud mist on the RGB three wave bands is, the larger the processing intensity difference on the RGB three wave bands is, and meanwhile, in a fog-free area, the RGB three wave bands are not processed;
according to the two rules, two adjusting coefficients J and K are introduced based on the initial transmissivity, and different defogging intensities are applied to the RGB three bands by utilizing the two coefficients;
D∈[0,11； DR＝D
J＝(0.9+0.1*D)^2； DG＝J*D
K＝(0·7+0·3*D)^2； DB(K) D formula 13
In equation 13, D is the initial transmittance obtained from the dark channel prior, and D isR、DG、DBThe transmittance for three bands of RGB is respectively, and different transmittance values represent different processing intensities for the three bands;
the cloud on the red band can be completely removed by the transmittance obtained by the initial dark channel prior method, so that the initial transmittance is taken as the transmittance on the red band. Then, by introducing adjusting coefficients J and K, the transmissivity on a green wave band and a blue wave band is calculated respectively based on the transmissivity calculated by an initial dark primary color prior method, and the processing intensity of the green wave band and the blue wave band is increased.
Compared with the prior art, the invention has the advantages and innovation points that:
firstly, a set of remote sensing image cloud and fog detection and removal system method is provided, and systematic scheme design and summary are made on the wave band self-adaptive demisting optimization processing method of a single remote sensing image. The optimized dark channel prior method can obtain satisfactory visual effect after demisting the remote sensing image, and effectively avoids the problem that the image is abnormally blue after demisting. By optimizing and improving the initial dark channel prior method in the two aspects, the cloud and fog pollution in the remote sensing image can be effectively removed, the spectral information of ground objects can be effectively maintained, and a satisfactory defogging effect is obtained in general.
Secondly, aiming at the problem that when the initial dark channel prior method is applied to the remote sensing image defogging, the high-brightness ground object is abnormal and dark, the invention provides the method which needs to carry out the self-adaptive defogging treatment on the high-brightness ground object in the remote sensing image. In consideration of the fact that the high-brightness ground object region in the remote sensing image does not accord with the dark primary color prior law, the method is based on the initial dark primary color prior demisting method, and when demisting treatment is carried out on the remote sensing image, the high-brightness ground objects in the image are firstly extracted, then adaptive demisting treatment with appropriate degree is carried out on the high-brightness ground objects, and the problem that the high-brightness ground objects are abnormally dark after demisting is avoided.
Thirdly, aiming at the problem that when the initial dark channel prior method is applied to defogging of the remote sensing image, the image is abnormally bluish, the invention provides the method which needs to perform self-adaptive defogging treatment on each wave band in the remote sensing image. Considering that different degrees of cloud and mist pollution of all wave bands in the remote sensing image are different, based on an initial dark channel prior demisting method, the invention sets corresponding demisting intensity for all the wave bands of the image, removes the cloud and mist pollution of all the wave bands in a self-adaptive manner, recovers a real and natural fog-free image and obtains a more real and natural visual effect.
According to the three aspects, the method creatively applies the dark primary color prior method originally used for defogging the outdoor image to the defogging of the remote sensing image according to the characteristics of the remote sensing image, and the defogging visual result and the experimental evaluation also prove the effectiveness of the self-adaptive dark primary color prior defogging method for the single remote sensing image.
Drawings
FIG. 1 is a flow chart of the band-adaptive defogging optimization processing method for a single remote sensing image according to the present invention.
FIG. 2 is a schematic diagram of the extraction result of high-brightness ground features in the remote sensing image.
FIG. 3 is a graph comparing the relationship between the parameter h and the process intensity in the adaptive defogging process of the present invention.
FIG. 4 is a schematic diagram comparing defogging effect diagrams of the optimized dark channel prior method of the present invention.
Detailed Description
The technical solution of the method for band-adaptive defogging optimization processing of a single remote sensing image provided by the present invention is further described below with reference to the accompanying drawings, so that those skilled in the art can better understand the present invention and can implement the same.
The cloud and fog in the remote sensing image are distributed uniformly in the image, and the thickness change in a local area is relatively smooth.
The invention provides a wave band self-adaptive demisting optimization processing method for a single remote sensing image, which firstly records the implementation process of a remote sensing image dark primary color prior demisting method in detail, aiming at the problem that cloud and fog pollution in an image can be basically removed after the remote sensing image is demisted by an initial dark primary color prior method, but the demisted image has two more serious problems, and the two more serious problems are:
the first problem is that the high-brightness ground objects are not in accordance with the assumed condition of dark channel prior, and the processing degree of the ground objects in the defogging process is too large, so that after the defogging process, the high-brightness ground objects are obviously abnormal and darker, and the spectral information of the corresponding ground objects is obviously abnormal.
The second problem is that the initial dark channel prior method is directed to a common natural image containing only three visible light bands of red, green and blue, while the remote sensing image usually has many bands. Meanwhile, the pollution degree of fog pollution to each wave band is different, fog generally only pollutes three visible light wave bands of red, green and blue, and the fog is hardly interfered by fog for a long wave band with the wavelength more than or equal to near infrared, so that a dark primary color prior method is utilized to generally only demist the three visible light wave bands of red, green and blue in a remote sensing image, and the demisting treatment is not carried out on the long wave bands of the near infrared wave band and the like. For the cloud and mist with larger concentration, the pollution degree of the cloud and mist with larger concentration to red, green and blue wave bands is different, the longer the wavelength is, the smaller the pollution degree of the cloud and mist is, and the shorter the wavelength is, the larger the pollution degree of the cloud and mist is. Therefore, the pollution degree of the blue wave band by fog is maximum, the pollution degree of the green wave band is second, and the pollution degree of the red wave band by fog is minimum, but the initial dark primary color prior method does not take the point into consideration, and the cloud and fog pollution on the green wave band and the blue wave band can not be completely removed in the demisted image due to the fact that the cloud and fog pollution on the green wave band and the blue wave band is processed in the same degree, and therefore the demisted image is extremely blue.
The method is improved on the basis of the remote sensing image initial dark channel prior demisting method aiming at the two problems. On one hand, in order to solve the problem that the high-brightness ground object in the image is abnormally dark after defogging by using an initial dark primary color prior method, the invention provides a processing idea that: firstly, the spectral information is utilized to extract high-brightness ground objects in the image, and then the extracted high-brightness ground objects are independently subjected to self-adaptive processing, so that a satisfactory experimental effect can be obtained, and the spectral information of the high-brightness ground objects is effectively reserved. On the other hand, after the reason that the image is abnormally bluish after defogging treatment by the initial dark primary color prior method is clarified, the invention provides that treatment of different degrees needs to be carried out on each wave band in the remote sensing image based on the initial dark primary color prior method, defogging treatment of the blue wave band is carried out to the maximum extent, defogging treatment of the green wave band is carried out to the second extent, defogging treatment of the red wave band is carried out to the smaller extent, and defogging treatment is not carried out on the long wave band of which the near infrared and the wavelength are larger than the near infrared wave. Experimental results prove that the optimized dark channel prior method can obtain a satisfactory visual effect after the remote sensing image is demisted, and effectively avoids the problem that the demisted image is abnormally blue. By optimizing and improving the initial dark channel prior method in the two aspects, the cloud and fog pollution in the remote sensing image can be effectively removed, the spectral information of ground objects can be effectively maintained, and a satisfactory defogging effect is obtained in general.
First, the procedure
The flow chart of the wave band self-adaptive demisting optimization processing method for a single remote sensing image provided by the invention is shown in figure 1, and comprises the following steps:
firstly, obtaining an initial remote sensing image;
secondly, carrying out standardized processing on the remote sensing image;
thirdly, calculating dark primary colors;
fourthly, generating a dark primary color image;
fifthly, calculating atmospheric light;
sixthly, calculating initial color transmittance;
seventhly, performing high-brightness ground object self-adaptive demisting treatment and wave band self-adaptive demisting treatment;
eighthly, optimizing the transmissivity of the guide image filtering drive;
and ninthly, solving the remote sensing image after defogging.
Second, demisting of initial dark channel prior of remote sensing image
The defogging method in the prior art ignores the characteristic that fog in a real remote sensing image is unevenly distributed, and performs defogging treatment on the whole image in an integral unified treatment mode, so that the cloud and fog in partial areas of the defogged image cannot be completely removed, and the problem of spectral distortion is caused by over-treatment in other areas.
The content of the initial dark channel prior demisting method is as follows: for most outdoor fog-free remote sensing images, a sky high brightness area is removed, a plurality of pixels always exist in a local area of the image, the intensity values of the pixels on one or a plurality of color channels of RGB (red, green and blue) three color channels are very low and are close to 0, namely, a local area is arbitrarily taken on the outdoor fog-free image, the minimum value on R, G, B color channels in the local area is counted, and the value is close to 0.
Randomly selecting an outdoor fog-free image A, and counting dark primary color values A in a local area of the fog-free image AdarkValue of dark primary color AdarkRepresented by formula 1:
in the above formula, AcOne color channel representing the image, r, g, b are color channels, and Ω (x) represents a square local area of the image centered on a pixel located at x.
For most outdoor fog-free images, except for areas where the sky itself is high in brightness, the locally obtained dark primary color intensity value in the image is very low and is close to 0, and equation 2 describes that:
Adark→ 0 formula 2
For outdoor fog-free images, the reasons for contributing to dark channel priors are mainly three: firstly, shadow areas often exist in outdoor fog-free remote sensing images, and the intensity values of pixels corresponding to the shadow areas on RGB three color channels are often very low; secondly, objects with obvious colors are usually found in outdoor fog-free images, and the values of the objects on a certain channel are high, while the values on other channels are low; third, there are usually dark objects in outdoor fog-free images, and the intensity values of these objects on the three RGB color channels are often very low.
Describing the composition of the fog image by using a structural equation of the fog image, wherein the equation is expressed in the form of formula 3:
b (x) ═ a (x) d (x) + E [1-d (x) ] formula 3
In the above formula, b (x) represents the observed foggy image in reality, a (x) is the ideal foggy scene and the desired final recovery result, E is the atmospheric light value in the global range and represents the intensity value of the pixel with the maximum fog concentration in the whole image, and d (x) represents the transmittance of the image, i.e. the proportion of the light which is not scattered by the atmosphere in the transmission process and finally enters the imaging system to the total light.
The structural equation of the fog image shows that a fog remote sensing image is composed of two parts, one part is a real fog-free scene, the other part is interference of atmospheric environment light, and the weights of the two parts are d (x) and 1-d (x) respectively.
The atmospheric light E represents the pixel value of the point with the maximum fog concentration in the whole image, and is obtained according to the following two steps:
step 1, taking 15 multiplied by 15 as a calculation step length to obtain a dark primary color image of an initial image, and selecting a front one thousandth brightest pixel in the dark primary color image;
and 2, corresponding one thousandth of brightest pixels to the initial image, selecting the pixels with the maximum brightness values, and taking the RGB values of the pixels with the maximum brightness values as the values of the atmospheric light E.
Assuming that the atmospheric light E is obtained, B represents the observed foggy image in reality, A is an ideal fogless scene, and AcRepresenting one color channel of an image, r, g, b are color channels, Ω (x) represents a square local area of the image centered on a pixel located at x, and in the local area Ω (x), the transmittance d' (x) is a constant value, and equation 4 can be found by combining equation 1 and equation 3:
combining with the dark primary color prior theory, in the outdoor fog-free image A, the dark primary color A of the local areadarkIs close to 0, while the atmospheric light value E in the global range is constantly positive, according to the above two conditions, equation 5 can be obtained:
the value of the transmittance d' (x) is obtained by substituting equation 5 into equation 4, as shown in equation 6:
after the normalization process is performed on the initial image,
i.e. the dark primary color values representing local areas in the image.
Finally, after obtaining the values of the atmospheric light C and the transmittance d' (x), and optimizing the initial transmittance map, the fog-free image a can be recovered by integrating equation 3 and equation 6:
through the steps, the dark channel prior theory is combined with the structural equation of the fog image, and the fog pollution in the remote sensing image is removed.
However, the initial dark channel prior has a great limitation, and when a scene in a remote sensing image is very close to atmospheric light, the dark channel prior fails because the dark channel prior cannot be close to 0 because of the dark channel value of an object with high brightness in the image, and if the initial dark channel prior method is used, the problem of over-processing occurs, so that the dark channel prior theory is not suitable for the scenes.
Problem when processing remote sensing image by initial dark channel prior demisting method
(I) abnormal dark problem of high-brightness ground object
After the remote sensing image is demisted by using the initial dark channel prior, cloud and fog pollution in the image can be basically removed, but the demisting of high-brightness ground objects such as bare soil in the image has the obvious problem of abnormal yellow and dark.
The reason for this problem is that the features in the scene are themselves high in brightness, and the dark channel prior theory is not applicable to these features when the spectral information is close to atmospheric light. Therefore, if these areas are defogged by the original dark process, the problem of over-treatment occurs.
(II) demisting image abnormal bluish problem
The common natural image only comprises red, green and blue visible light wave bands generally, while the wave band information of part of remote sensing images is rich, and the common natural image also comprises near infrared wave bands, middle infrared wave bands and far infrared wave bands besides the red, green and blue visible light wave bands. And as for the near infrared band and the intermediate infrared band and the thermal infrared band with the wavelength larger than the near infrared wavelength, the influence of fog is hardly caused, so that when the multi-band remote sensing image is demisted, demisting treatment is only carried out on the three visible light bands of RGB.
The image subjected to defogging treatment by using the initial dark channel prior method has the problem of obvious bluish color, and the reason for generating the abnormal bluish color is that the influence degrees of fog on various wave bands are different for multiband images, and the influence degrees of fog on various wave bands are gradually reduced along with the increase of the wavelength. Therefore, the influence degrees of the fog on the red, green and blue wave bands are different, the influence degree of the fog on the blue wave band is the largest, the influence degree on the green wave band is the next, and the influence degree on the red wave band is the smallest.
However, the initial dark channel prior method does not take this point into consideration, and the same transmittance d is used for the three visible light bands of RGB to perform the same defogging treatment, so that cloud and mist pollution on the red band after defogging can be completely removed, and cloud and mist pollution on the green band and the blue band can not be completely removed, so that the image after defogging has an abnormal blue tendency.
The invention aims at the problems of high-brightness ground object in the demisted image being abnormal and dark and the demisted image being abnormal and blue, and carries out self-adaptive demisting optimization treatment, and the initial dark primary color prior demisting method is expected to be applied to the demisting of the remote sensing image to obtain satisfactory demisting effect.
Four, high brightness ground object self-adaptive demisting treatment method
In order to solve the problem that high-brightness ground objects in the demisted image are abnormal and dark, the invention provides a self-adaptive demisting processing method for the high-brightness ground objects, which comprises the following steps:
firstly, extracting high-brightness ground objects in a remote sensing image according to different spectral characteristics of the ground objects; then, the extracted high-brightness ground object is subjected to self-adaptive defogging treatment, and appropriate defogging intensity is given.
(1) Extraction of high-brightness ground features
If the image only has three wave bands of RGB, according to the difference of spectral characteristics of each object, adopting a discrimination condition F to extract high-brightness ground objects in the remote sensing image:
in formula 8, RGB represents three bands of red, green, and blue of the image, respectively.
Judging each pixel in the remote sensing image by using the judgment condition of the formula 8, and judging the pixel as a high-brightness ground object if the F value corresponding to the pixel is less than 1; if the F value is more than 1.5, judging the ground object as non-high-brightness ground object; and if the F value is between 1 and 1.5, judging that the transition from the high-brightness ground object to the non-high-brightness ground object is formed. Fig. 2 is a schematic diagram of the extraction result of the high-brightness ground object in the remote sensing image, and the result of the extraction of the high-brightness ground object by using the discrimination condition is shown in the upper diagram as an initial image and in the lower diagram as the extraction result of the high-brightness ground object, and most of the high-brightness ground objects can be accurately extracted.
If the remote sensing image has a plurality of wave bands such as near infrared besides three wave bands of red, green and blue, the high-brightness ground object formula in the image is accurately extracted by using two coefficients of the normalized water index NDWI and the normalized vegetation index NDVI as judgment conditions,
in formulas 9 and 10, Nir represents a near infrared band, R represents a red band, and G represents a green band.
In a remote sensing image of a specific embodiment, high-brightness ground objects in the image can be accurately extracted by using NDVI <0.2& NDWI <0 as a judgment condition.
(2) Self-adaptive demisting treatment for high-brightness ground objects
And after extracting the high-brightness ground objects in the remote sensing image, carrying out defogging treatment, giving appropriate defogging intensity to the high-brightness ground objects, and carrying out self-adaptive defogging treatment. A parameter h is introduced in the formula 7 to adjust the defogging treatment intensity, the value range of h is between 0 and 1, d' (x) is the corrected transmissivity,
FIG. 3 is a comparison graph of the relationship between the parameter h and the processing intensity in the adaptive defogging process, wherein h is respectively assigned to 0.6, 0.8 and 1, and as can be seen from the comparison between the defogging results, the larger the h value is, the stronger the processing intensity is; the smaller the value of h, the weaker the treatment strength. And by adjusting the h value, the extracted high-brightness ground object is endowed with smaller processing intensity, namely the h value corresponding to the high-brightness ground object is smaller, the h value corresponding to the non-high-brightness ground object is larger and is between the high-brightness ground object and the non-high-brightness ground object, and the h value is gradually changed.
For an image with three RGB bands, the value of h is shown in formula 12:
and correcting the transmissivity of the high-brightness ground area by introducing a parameter h, and adjusting the processing intensity of the high-brightness ground object in the remote sensing image. And performing smooth filtering treatment on the corrected transmittance graph to obtain a final transmittance graph, wherein the smooth filtering treatment method adopted by the invention is guide graph filtering, and boundary information among different ground objects can be effectively reserved while the remote sensing image is subjected to smooth filtering treatment by adopting the guide graph filtering.
The atmospheric light E and the corrected transmittance d' (x) are solved, and then the image is finally subjected to defogging processing according to equation 11.
Fig. 4 is a schematic diagram showing a comparison of defogging effects of the optimized dark channel prior method, which respectively shows a defogging result of the initial dark channel prior method and a defogging result of the optimized adaptive dark channel prior method, where (a) is an initial remote sensing image, (b) is a defogging result of the initial dark channel method, (c) is an extracted high-brightness feature, and (d) is a defogging result of the optimized adaptive dark channel method. Experimental results show that after the high-brightness ground object in the image is subjected to self-adaptive processing by using the optimized dark primary color prior method, the interference of cloud and fog in the image can be removed, the spectral information of the high-brightness ground object can be effectively reserved, and the problem of abnormal darkness caused by over-processing when the initial dark primary color prior method is used for processing the high-brightness ground object is avoided.
Five-wave band self-adaptive demisting treatment
The influence degree of cloud and fog pollution on each wave band in the remote sensing image is different, and the influence degree of fog on each wave band is gradually reduced along with the increase of the wavelength on the multiband image. And for the near infrared band and the band with the wavelength larger than the near infrared wavelength, such as the middle infrared band and the thermal infrared band, the influence of fog is hardly generated.
Therefore, when defogging is carried out on the multiband remote sensing image, defogging treatment is only carried out on the three visible wave bands of RGB. Meanwhile, in the initial dark channel prior method, the same degree of defogging treatment is performed on the three bands of RGB by using the same transmittance d, which may cause that cloud fog on the green band and the blue band cannot be completely removed, resulting in that the defogged image is bluish.
In order to solve the problem that the image after demisting is abnormally bluish, the invention provides a model: after the transmissivity is calculated through the initial dark primary color, the transmissivity aiming at each wave band of RGB is calculated, and self-adaptive demisting treatment with different degrees is carried out on the three wave bands of RGB.
Meanwhile, the distribution concentration of the cloud mist is assumed, and the following two conditions are met on three RGB wave bands:
firstly, to the cloud and fog of the same concentration, to red green blue three wave band's influence degree increase progressively, the transmissivity of the three wave bands of RGB ranks as: the red band has the highest transmittance, the green band is the second order, and the blue band is the lowest; therefore, in the foggy area, the processing intensity of the RGB three wave bands is increased progressively;
secondly, the more dense the concentration of the cloud and fog, the larger the difference of the influence of the cloud and fog on the three RGB wave bands is, the larger the difference of the processing intensity on the three RGB wave bands is, and meanwhile, in a fog-free area, the three RGB wave bands are not processed.
According to the above two rules, two adjustment coefficients J and K are introduced based on the initial transmittance, and different defogging intensities are applied to the RGB three bands using the two coefficients.
D∈[0,1]； DR＝D
J＝(0.9+0.1*D)^2； DG＝J*D
K＝(0·7+0·3*D)^2； DB＝K*DFormula l3
In equation 13, D is the initial transmittance obtained from the dark channel prior, and D isR、DG、DBThe transmittance for the three RGB bands, respectively, different transmittance values represent different processing intensities for the three bands.
The cloud on the red band can be completely removed by the transmittance obtained by the initial dark channel prior method, so that the initial transmittance is taken as the transmittance on the red band. Then, by introducing adjusting coefficients J and K and calculating the transmissivity on the green wave band and the blue wave band respectively based on the transmissivity calculated by the initial dark primary color prior method, the processing intensity on the green wave band and the blue wave band is increased, and the problem that the image is abnormal and blue due to the fact that cloud mist on the blue-green wave band cannot be completely removed when the initial dark primary color prior method is used for demisting is avoided.
Experiments show that after the optimized dark primary color prior method is used for carrying out adaptive processing on various wave bands in the remote sensing image in different degrees, cloud interference in the image can be removed, the spectral information of the initial image can be effectively reserved, and the problem that the initial dark primary color prior method is abnormal and blue when the remote sensing image is processed is solved.
Sixthly, transmissivity optimization driven by guide diagram filtering
Through the steps, the initial transmittance corresponding to the original remote sensing image can be obtained, the initial transmittance can probably reflect the cloud distribution condition in the image, but the initial transmittance has the problem of local agglomeration because a square area of 15 x 15 is adopted when the dark primary color value is counted before. In order to obtain a more accurate and smoother transmittance map, the initial transmittance needs to be filtered.
The filtering method adopted by the invention is guide map filtering, which not only realizes smooth filtering effect, but also can keep the edge information of the image, and has high processing efficiency and higher running speed.
The core framework of the directed graph filtering is: a specified point on the function and a point in the vicinity of the specified point are represented by a linear relation, and a complex high-level function can be represented by a plurality of local linear function synthesis.
The remote sensing image processing driven by the guide map filtering comprises an input image L, a guide image M and an output image N, wherein the input image L and the reference guide image M can be the same image or different images.
The guide graph filtering assumes that an image can be regarded as a two-dimensional function, and the filtered smoothed image N and the input image L satisfy a linear relationship within a two-dimensional window, which is expressed by equation 14:
in formula 14, wkRepresents a square window, M is the value of the guide image, N is the value of the output image, k represents the index number of the window, i represents the pixel number of the input and output images, akAnd bkRepresenting the coefficients of the guide map filter corresponding to the linear function when the center position of the filter window is located at k, equation 14 indicates that there is a linear relationship between the output image L and the input image N in a local area.
The input image L is generally an image to be processed, and the guide image M may be another image or the image to be processed itself.
The gradient is found for both sides of equation 14, which yields the following results:
in equation 15, when the reference guide image M has specific gradient information, the output image N after the guide map filtering process also has similar gradient information, and therefore the output image N has similar edge information to the reference image M, which is why the guide map filtering process can maintain the edge characteristics at the same time as the smoothing filtering process.
The key of the guide graph filtering lies in solving the linear coefficient a in the local areakAnd bkTo minimize the difference between the input image and the output image, i.e. toThe following function values are guaranteed to be minimal:
in equation 16, L is the input image, M is the guide image, and e is a regularization parameter for preventing smoothing
Too large, e is also an important parameter for adjusting the filtering effect.
By means of linear regression analysis, akAnd bkThe specific expression of the optimal solution is as follows:
in formulae 17 and 18, wkIs represented in a local window wkThe number of pixels in the interior is,
representing the guide image M in the local window wkVariance of (1), μkRepresenting the guide image in a local window wkThe average value of the gray levels of the pixels in (b),
indicating that the image to be processed is in the window wkAverage value of the gray levels of the middle pixels.
In calculating each local window wkThe corresponding linear coefficient is often cross-covered by a plurality of different windows for a particular pixel, i.e. each pixel is represented by a plurality of linear functions. Therefore, according to the core framework of the guide map filtering described above, all linear functions including the pixel are averaged to obtain an output value corresponding to the pixel. The obtained output value can be expressed by the following equation：
For the guide map filtering, the final filtering effect is mainly influenced by the window radius r and the regularization parameter e, the larger the filtering window radius r is, the better the smoothing effect of a local area is, and the smoother the output image is; the regularization parameter e is used for judging the degree of edge preservation, the larger e is, the smoother the output image is, and the less edge information in the image is, and the smaller e is, the more edge information in the initial image is preserved in the output image.
The self-adaptive dark primary color defogging technology for the single remote sensing image can obtain a better defogging effect, can effectively remove cloud and fog interference in the image, can effectively retain spectral information of the image, has good reliability, and obtains a very satisfactory defogging effect.
Claims (9)
1. The wave band self-adaptive demisting optimization processing method of a single remote sensing image is characterized in that improvement is made on the basis of an initial dark primary color prior demisting method,
on one hand, a processing method for solving the problem that high-brightness ground objects in an image are abnormally dark after defogging by using an initial dark primary color prior method is provided: adopting high-brightness ground object self-adaptive demisting treatment, firstly extracting high-brightness ground objects in a remote sensing image by using spectral information, and then independently carrying out self-adaptive treatment on the extracted high-brightness ground objects to keep the spectral information of the high-brightness ground objects;
on the other hand, a processing method for solving the problem that the image after defogging processing is abnormally blue by using an initial dark channel prior method is provided: adopting wave band self-adaptive demisting treatment, carrying out different degrees of treatment on each wave band in the remote sensing image, wherein the demisting treatment on a blue wave band is maximum, the demisting treatment on a red wave band is carried out next to a green wave band to a lesser extent, and the demisting treatment is not carried out on a long wave band of which the near infrared and wavelength are greater than the near infrared wave band;
the wave band self-adaptive demisting optimization processing method of the single remote sensing image comprises the following steps:
firstly, obtaining an initial remote sensing image;
secondly, carrying out standardized processing on the remote sensing image;
thirdly, calculating dark primary colors;
fourthly, generating a dark primary color image;
fifthly, calculating atmospheric light;
sixthly, calculating initial color transmittance;
seventhly, performing high-brightness ground object self-adaptive demisting treatment and wave band self-adaptive demisting treatment;
eighthly, optimizing the transmissivity of the guide image filtering drive;
and ninthly, solving the remote sensing image after defogging.
2. The method for band-adaptive defogging optimization processing of a single remote sensing image according to claim 1 is characterized in that the method is based on an initial dark primary color prior defogging method to remove unevenly distributed cloud and fog pollution in the remote sensing image;
the content of the initial dark channel prior demisting method is as follows: for most outdoor fog-free remote sensing images, a sky high brightness area is removed, a plurality of pixels always exist in a local area of the image, the intensity values of the pixels on one or a plurality of color channels of RGB (red, green and blue) three color channels are very low and are close to 0, namely, a local area is arbitrarily selected on the outdoor fog-free image, the minimum value on R, G, B color channels in the local area is counted, and the value is close to 0;
randomly selecting an outdoor fog-free image A, and counting dark primary color values A in a local area of the fog-free image AdarkValue of dark primary color AdarkRepresented by formula 1:
in the formula 1, AcOne color channel representing the image, r, g, b are color channels, Ω (x) represents a square local area of the image centered on a pixel located at x;
for most outdoor fog-free images, except for areas with high brightness of the sky, the locally obtained dark primary color intensity value in the image is very low and is close to 0, and formula 2 is described as follows:
Adark→ 0 formula 2
Describing the composition of the fog image by using a structural equation of the fog image, wherein the equation is expressed in the form of formula 3:
b (x) ═ a (x) d (x) + E [1-d (x) ] formula 3
In formula 3, b (x) represents a fog image observed in reality, a (x) is an ideal fog-free scene and is also a result of the desired final restoration, E is an atmospheric light value in the global range and represents an intensity value of a pixel with the maximum fog concentration in the whole image, d (x) represents the transmittance of the image, namely, the light which is not scattered by the atmosphere in the transmission process and finally enters the imaging system accounts for the proportion of the total light;
the structural equation of the fog image shows that a fog remote sensing image is composed of two parts, one part is a real fog-free scene, the other part is interference of atmospheric environment light, and the weights of the two parts are d (x) and 1-d (x) respectively.
3. The method for band-adaptive defogging optimization processing of a single remote sensing image according to claim 2, wherein the atmospheric light E represents the pixel value of the point with the maximum fog concentration in the whole image, and the atmospheric light E is obtained according to the following two steps:
step 1, taking 15 multiplied by 15 as a calculation step length to obtain a dark primary color image of an initial image, and selecting a front one thousandth brightest pixel in the dark primary color image;
and 2, corresponding one thousandth of brightest pixels to the initial image, selecting the pixels with the maximum brightness values, and taking the RGB values of the pixels with the maximum brightness values as the values of the atmospheric light E.
4. The method of claim 3, wherein, assuming that the atmospheric light E has been determined, B represents a fog image observed in reality, A is an ideal fog-free scene, and A is an ideal fog-free scenecRepresenting one color channel of an image, r, g, b are color channels, Ω (x) represents a square local area of the image centered on a pixel located at x, and in the local area Ω (x), the transmittance d' (x) is a constant value, and equation 4 can be found by combining equation 1 and equation 3:
5. the method for band-adaptive defogging optimization processing of a single remote sensing image according to claim 4, wherein a dark primary color A of a local area in the outdoor fog-free image A is combined with a dark primary color prior theorydarkIs close to 0, while the atmospheric light value E in the global range is constantly positive, according to the above two conditions, equation 5 can be obtained:
the value of the transmittance d' (x) is obtained by substituting equation 5 into equation 4, as shown in equation 6:
after the normalization process is performed on the initial image,
namely representing the dark primary color value of a local area in the image;
finally, after obtaining the values of the atmospheric light C and the transmittance d' (x), and optimizing the initial transmittance map, the fog-free image a can be recovered by integrating equation 3 and equation 6:
through the steps, the dark channel prior theory is combined with the structural equation of the fog image, and the fog pollution in the remote sensing image is removed.
6. The method for band-adaptive defogging and optimizing of a single remote sensing image according to claim 1 is characterized in that the invention provides a high-brightness ground object adaptive defogging method which comprises the following steps: firstly, extracting high-brightness ground objects in a remote sensing image according to different spectral characteristics of the ground objects; then, the extracted high-brightness ground object is subjected to self-adaptive defogging treatment, and appropriate defogging intensity is given.
7. The method for band-adaptive defogging optimization processing of a single remote sensing image according to claim 6, wherein in the extraction of the high-brightness ground object, if the image only has three bands of RGB, the high-brightness ground object in the remote sensing image is extracted by adopting a discrimination condition F according to the difference of spectral characteristics of each ground object:
in the formula 8, RGB represents three bands of red, green and blue of the image, respectively;
judging each pixel in the remote sensing image by using the judgment condition of the formula 8, and judging the pixel as a high-brightness ground object if the F value corresponding to the pixel is less than 1; if the F value is more than 1.5, judging the ground object as non-high-brightness ground object; if the F value is between 1 and 1.5, judging that the transition from the high-brightness ground object to the non-high-brightness ground object is carried out;
if the remote sensing image has a plurality of wave bands such as near infrared besides three wave bands of red, green and blue, the high-brightness ground object formula in the image is accurately extracted by using two coefficients of the normalized water index NDWI and the normalized vegetation index NDVI as judgment conditions,
in formulas 9 and 10, Nir represents a near infrared band, R represents a red band, and G represents a green band.
8. The method as claimed in claim 6, wherein the adaptive defogging process is performed on high brightness ground objects, the high brightness ground objects in the remote sensing image are extracted, then the defogging process is performed, the high brightness ground objects are endowed with proper defogging intensity, the adaptive defogging process is performed, a parameter h is introduced to adjust the defogging intensity, the value range of h is 0-1, d' (x) is the transmission rate after correction,
for an image with three RGB bands, the value of h is shown in formula 12:
the transmissivity of the high-brightness ground area is corrected by introducing a parameter h, the processing intensity of the high-brightness ground object in the remote sensing image is adjusted, and the corrected transmissivity image is subjected to smooth filtering processing to obtain a final transmissivity image;
the atmospheric light E and the corrected transmittance d' (x) are solved, and then the image is finally subjected to defogging processing according to equation 11.
9. The method for band-adaptive defogging and optimizing a single remote sensing image according to claim 1, wherein in the band-adaptive defogging process, only three visible bands of RGB are defogged, and the invention provides a model: after the transmissivity is calculated through the initial dark primary color, the transmissivity aiming at each wave band of RGB is calculated, and self-adaptive demisting treatment with different degrees is carried out on the three wave bands of RGB;
meanwhile, the distribution concentration of the cloud mist is assumed, and the following two conditions are met on three RGB wave bands:
firstly, to the cloud and fog of the same concentration, to red green blue three wave band's influence degree increase progressively, the transmissivity of the three wave bands of RGB ranks as: the red band has the highest transmittance, the green band is the second order, and the blue band is the lowest; therefore, in the foggy area, the processing intensity of the RGB three wave bands is increased progressively;
secondly, the more dense the concentration of the cloud mist, the larger the influence difference of the cloud mist on the RGB three wave bands is, the larger the processing intensity difference on the RGB three wave bands is, and meanwhile, in a fog-free area, the RGB three wave bands are not processed;
according to the two rules, two adjusting coefficients J and K are introduced based on the initial transmissivity, and different defogging intensities are applied to the RGB three bands by utilizing the two coefficients;
D∈[0，1]；DR＝D
J＝(0.9+0.1*D)^2；DG＝J*D
K＝(0.7+0.3*D)^2；DBk x D formula l3
In equation 13, D is the initial transmittance obtained from the dark channel prior, and D isR、DG、DBThe transmittance for three bands of RGB is respectively, and different transmittance values represent different processing intensities for the three bands;
the cloud on the red band can be completely removed by the transmittance obtained by the initial dark channel prior method, so that the initial transmittance is taken as the transmittance on the red band. Then, by introducing adjusting coefficients J and K, the transmissivity on a green wave band and a blue wave band is calculated respectively based on the transmissivity calculated by an initial dark primary color prior method, and the processing intensity of the green wave band and the blue wave band is increased.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010345923.6A CN111539891A (en) | 2020-04-27 | 2020-04-27 | Wave band self-adaptive demisting optimization processing method for single remote sensing image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010345923.6A CN111539891A (en) | 2020-04-27 | 2020-04-27 | Wave band self-adaptive demisting optimization processing method for single remote sensing image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111539891A true CN111539891A (en) | 2020-08-14 |
Family
ID=71967534
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010345923.6A Pending CN111539891A (en) | 2020-04-27 | 2020-04-27 | Wave band self-adaptive demisting optimization processing method for single remote sensing image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111539891A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113838065A (en) * | 2021-09-23 | 2021-12-24 | 江苏天汇空间信息研究院有限公司 | Automatic cloud removing method based on image markers |
Citations (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080317287A1 (en) * | 2007-06-13 | 2008-12-25 | Denso Corporation | Image processing apparatus for reducing effects of fog on images obtained by vehicle-mounted camera and driver support apparatus which utilizies resultant processed images |
| JP2011193381A (en) * | 2010-03-16 | 2011-09-29 | Central Glass Co Ltd | Plane antenna |
| CN102567977A (en) * | 2011-12-31 | 2012-07-11 | 南京理工大学 | Self-adaptive fusing method of infrared polarization image based on wavelets |
| WO2014018213A1 (en) * | 2012-07-23 | 2014-01-30 | Lockheed Martin Corporation | Realtime long range imaging scatter reduction |
| CN103824260A (en) * | 2013-12-20 | 2014-05-28 | 中国科学院遥感与数字地球研究所 | High-resolution remote sensing image mist rapid removal technology |
| JP2015031564A (en) * | 2013-08-01 | 2015-02-16 | 株式会社リコー | Deposit detection device, and equipment control system for transfer device |
| US20150279011A1 (en) * | 2014-03-27 | 2015-10-01 | Samsung Techwin Co., Ltd. | Defogging system and defogging method |
| CN104966273A (en) * | 2015-05-29 | 2015-10-07 | 昆山数字城市信息技术有限公司 | DCM-HTM haze-removing method suitably used for optical remote sensing images |
| CN105405112A (en) * | 2015-12-29 | 2016-03-16 | 中国人民解放军信息工程大学 | Multispectral satellite image distance deviation index defogging method |
| CN105913387A (en) * | 2016-04-01 | 2016-08-31 | 浙江大学 | Visible light and near-infrared multiband remote sensing image defogging method based on Angstrom index |
| US20160328832A1 (en) * | 2015-05-08 | 2016-11-10 | Hanwha Techwin Co., Ltd. | Defog system and defog method |
| CN106709876A (en) * | 2015-08-27 | 2017-05-24 | 上海沃韦信息科技有限公司 | Optical remote sensing image defogging method based on the principle of dark pixel |
| CN106709893A (en) * | 2016-12-28 | 2017-05-24 | 西北大学 | All-time haze image sharpness recovery method |
| CN106780384A (en) * | 2016-12-15 | 2017-05-31 | 上海交通大学 | A kind of the real-time of cold light source abdominal cavity image parameters self adaptation that be applicable goes smog method |
| CN107038691A (en) * | 2017-04-12 | 2017-08-11 | 北京爱特拉斯信息科技有限公司 | The dark primary remote sensing image defogging method aided in based on cloud detection |
| CN107256541A (en) * | 2017-06-15 | 2017-10-17 | 北京航空航天大学 | A kind of multi-spectral remote sensing image defogging method based on convolutional neural networks |
| CN107317968A (en) * | 2017-07-27 | 2017-11-03 | 广东欧珀移动通信有限公司 | Image defogging method, device, computer can storage medium and mobile terminals |
| CN107317969A (en) * | 2017-07-27 | 2017-11-03 | 广东欧珀移动通信有限公司 | Image defogging method, device, computer can storage medium and mobile terminals |
| CN107424198A (en) * | 2017-07-27 | 2017-12-01 | 广东欧珀移动通信有限公司 | Image processing method, device, mobile terminal and computer-readable recording medium |
| CN107578386A (en) * | 2017-09-01 | 2018-01-12 | 荆门程远电子科技有限公司 | A kind of optimization defogging processing method of unmanned plane shooting image |
| CN107636724A (en) * | 2015-07-17 | 2018-01-26 | 中央大学校产学协力团 | The adaptive demister of wavelength and method of image |
| CN108537744A (en) * | 2018-03-14 | 2018-09-14 | 北京师范大学 | A kind of coloured image luminance component homomorphic filtering defogging method |
| CN109035150A (en) * | 2018-04-24 | 2018-12-18 | 湖南省有色地质勘查研究院 | A kind of remote sensing image data defogging method based on middle-low alloy steels |
| WO2019017246A1 (en) * | 2017-07-18 | 2019-01-24 | Ａｇｃ株式会社 | Window glass for vehicles |
| CN109883966A (en) * | 2019-02-26 | 2019-06-14 | 江苏大学 | A method of Eriocheir sinensis amount of cure is detected based on multispectral image technology |
| WO2019205707A1 (en) * | 2018-04-26 | 2019-10-31 | 长安大学 | Dark channel based image defogging method for linear self-adaptive improvement of global atmospheric light |
| CN110428008A (en) * | 2019-08-02 | 2019-11-08 | 深圳市唯特视科技有限公司 | A kind of target detection and identification device and method based on more merge sensors |
| CN110738622A (en) * | 2019-10-17 | 2020-01-31 | 温州大学 | Lightweight neural network single image defogging method based on multi-scale convolution |
-
2020
- 2020-04-27 CN CN202010345923.6A patent/CN111539891A/en active Pending
Patent Citations (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080317287A1 (en) * | 2007-06-13 | 2008-12-25 | Denso Corporation | Image processing apparatus for reducing effects of fog on images obtained by vehicle-mounted camera and driver support apparatus which utilizies resultant processed images |
| JP2011193381A (en) * | 2010-03-16 | 2011-09-29 | Central Glass Co Ltd | Plane antenna |
| CN102567977A (en) * | 2011-12-31 | 2012-07-11 | 南京理工大学 | Self-adaptive fusing method of infrared polarization image based on wavelets |
| WO2014018213A1 (en) * | 2012-07-23 | 2014-01-30 | Lockheed Martin Corporation | Realtime long range imaging scatter reduction |
| JP2015031564A (en) * | 2013-08-01 | 2015-02-16 | 株式会社リコー | Deposit detection device, and equipment control system for transfer device |
| CN103824260A (en) * | 2013-12-20 | 2014-05-28 | 中国科学院遥感与数字地球研究所 | High-resolution remote sensing image mist rapid removal technology |
| US20150279011A1 (en) * | 2014-03-27 | 2015-10-01 | Samsung Techwin Co., Ltd. | Defogging system and defogging method |
| US20160328832A1 (en) * | 2015-05-08 | 2016-11-10 | Hanwha Techwin Co., Ltd. | Defog system and defog method |
| CN104966273A (en) * | 2015-05-29 | 2015-10-07 | 昆山数字城市信息技术有限公司 | DCM-HTM haze-removing method suitably used for optical remote sensing images |
| CN107636724A (en) * | 2015-07-17 | 2018-01-26 | 中央大学校产学协力团 | The adaptive demister of wavelength and method of image |
| CN106709876A (en) * | 2015-08-27 | 2017-05-24 | 上海沃韦信息科技有限公司 | Optical remote sensing image defogging method based on the principle of dark pixel |
| CN105405112A (en) * | 2015-12-29 | 2016-03-16 | 中国人民解放军信息工程大学 | Multispectral satellite image distance deviation index defogging method |
| CN105913387A (en) * | 2016-04-01 | 2016-08-31 | 浙江大学 | Visible light and near-infrared multiband remote sensing image defogging method based on Angstrom index |
| CN106780384A (en) * | 2016-12-15 | 2017-05-31 | 上海交通大学 | A kind of the real-time of cold light source abdominal cavity image parameters self adaptation that be applicable goes smog method |
| CN106709893A (en) * | 2016-12-28 | 2017-05-24 | 西北大学 | All-time haze image sharpness recovery method |
| CN107038691A (en) * | 2017-04-12 | 2017-08-11 | 北京爱特拉斯信息科技有限公司 | The dark primary remote sensing image defogging method aided in based on cloud detection |
| CN107256541A (en) * | 2017-06-15 | 2017-10-17 | 北京航空航天大学 | A kind of multi-spectral remote sensing image defogging method based on convolutional neural networks |
| WO2019017246A1 (en) * | 2017-07-18 | 2019-01-24 | Ａｇｃ株式会社 | Window glass for vehicles |
| CN107317968A (en) * | 2017-07-27 | 2017-11-03 | 广东欧珀移动通信有限公司 | Image defogging method, device, computer can storage medium and mobile terminals |
| CN107317969A (en) * | 2017-07-27 | 2017-11-03 | 广东欧珀移动通信有限公司 | Image defogging method, device, computer can storage medium and mobile terminals |
| CN107424198A (en) * | 2017-07-27 | 2017-12-01 | 广东欧珀移动通信有限公司 | Image processing method, device, mobile terminal and computer-readable recording medium |
| CN107578386A (en) * | 2017-09-01 | 2018-01-12 | 荆门程远电子科技有限公司 | A kind of optimization defogging processing method of unmanned plane shooting image |
| CN108537744A (en) * | 2018-03-14 | 2018-09-14 | 北京师范大学 | A kind of coloured image luminance component homomorphic filtering defogging method |
| CN109035150A (en) * | 2018-04-24 | 2018-12-18 | 湖南省有色地质勘查研究院 | A kind of remote sensing image data defogging method based on middle-low alloy steels |
| WO2019205707A1 (en) * | 2018-04-26 | 2019-10-31 | 长安大学 | Dark channel based image defogging method for linear self-adaptive improvement of global atmospheric light |
| CN109883966A (en) * | 2019-02-26 | 2019-06-14 | 江苏大学 | A method of Eriocheir sinensis amount of cure is detected based on multispectral image technology |
| CN110428008A (en) * | 2019-08-02 | 2019-11-08 | 深圳市唯特视科技有限公司 | A kind of target detection and identification device and method based on more merge sensors |
| CN110738622A (en) * | 2019-10-17 | 2020-01-31 | 温州大学 | Lightweight neural network single image defogging method based on multi-scale convolution |
Non-Patent Citations (3)
| Title |
|---|
| FAN, XUNLI等: "Image defogging approach based on incident light frequency", 《MULTIMEDIA TOOLS AND APPLICATIONS》, vol. 78, no. 13, 10 January 2019 (2019-01-10), pages 17653 - 17672, XP036832561, DOI: 10.1007/s11042-018-7103-1 * |
| 袁全: "单幅遥感影像的自适应暗原色去雾技术研究", 《武汉大学机构知识库》, 31 December 2018 (2018-12-31), pages 22 - 42 * |
| 赵阳等: "基于自适应改进的遥感图像去雾算法研究", 《电子设计工程》, vol. 27, no. 19, 31 October 2019 (2019-10-31), pages 164 - 169 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113838065A (en) * | 2021-09-23 | 2021-12-24 | 江苏天汇空间信息研究院有限公司 | Automatic cloud removing method based on image markers |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108596849B (en) | Single image defogging method based on sky region segmentation | |
| CN106157267B (en) | Image defogging transmissivity optimization method based on dark channel prior | |
| CN111292258B (en) | Image defogging method based on dark channel prior and bright channel prior | |
| CN101783012B (en) | Automatic image defogging method based on dark primary colour | |
| Gao et al. | Sand-dust image restoration based on reversing the blue channel prior | |
| CN108389175B (en) | Image defogging method integrating variation function and color attenuation prior | |
| CN111667433B (en) | Unmanned aerial vehicle image defogging method based on simple linear iterative clustering optimization | |
| CN109087254B (en) | Unmanned aerial vehicle aerial image haze sky and white area self-adaptive processing method | |
| CN108269244B (en) | Image defogging system based on deep learning and prior constraint | |
| CN107798670B (en) | Dark channel prior image defogging method using image guide filter | |
| CN102306384A (en) | Color constancy processing method based on single image | |
| CN111161167B (en) | Single image defogging method based on middle channel compensation and self-adaptive atmospheric light estimation | |
| CN108564543A (en) | A kind of underwater picture color compensation method based on electromagnetic theory | |
| CN110827221A (en) | Single image defogging method based on double-channel prior and side window guide filtering | |
| Bi et al. | Haze removal for a single remote sensing image using low-rank and sparse prior | |
| CN106971380A (en) | A kind of contrast enhancing and application of the visual saliency optimization method in golf course figure | |
| CN106846258A (en) | A kind of single image to the fog method based on weighted least squares filtering | |
| CN114359143A (en) | Remote sensing image road extraction method | |
| CN111325688B (en) | Unmanned aerial vehicle image defogging method for optimizing atmosphere light by fusion morphology clustering | |
| Kumar et al. | Enhancing scene perception using a multispectral fusion of visible–near‐infrared image pair | |
| CN109118450A (en) | A kind of low-quality images Enhancement Method under the conditions of dust and sand weather | |
| CN109003238A (en) | A kind of image haze minimizing technology based on model and histogram and grey level enhancement | |
| CN111539891A (en) | Wave band self-adaptive demisting optimization processing method for single remote sensing image | |
| Liang et al. | Learning to remove sandstorm for image enhancement | |
| CN111311502A (en) | Method for processing foggy day image by using bidirectional weighted fusion |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |