Remote sensing image defogging method based on minimum entropy optimization

Abstract

The invention provides a minimum entropy optimization-based remote sensing image defogging method, which comprises the following steps of: firstly, acquiring a fog-containing image in haze weather, then carrying out relative total variation smoothing on the fog-containing image to obtain a main body structure diagram, carrying out brightness mapping on the main body structure diagram in horizontal and vertical directions, and carrying out defogging on the main body structure diagram; taking a row and a column with the maximum brightness mapping sum of the adjacent areas and the adjacent areas as atmospheric light candidate areas, and taking a row and a column with the minimum information entropy in the candidate areas as a scene atmospheric light position; and finally, obtaining the haze transmissivity of the image according to the haze diffusion length and the clustering result of the unit spherical coordinate system, and outputting a clear image according to an atmospheric scattering model. Compared with a traditional method, the method is more accurate in selection of the scene atmosphere light value and the haze transmissivity of the fog-containing image, the restored image is small in halation effect, and the method is of great significance to sharpness of the fog-containing image.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a remote sensing image defogging method based on minimum entropy optimization. Background Art

[0002] With the rapid development of image processing, using remote sensing images to obtain scene information has become an effective means. However, haze in the atmosphere can cause color distortion and contrast reduction of objects in the image, making remote sensing images blurry and dim in foggy conditions. It also increases the problem of uneven lighting. This interference makes it difficult for computer vision systems to accurately segment and identify target objects in the image. Secondly, haze can also cause the loss of depth information of objects in the image. The depth difference is very small compared to nearby objects, making it difficult for computer vision systems to accurately understand scene information.

[0003] In recent years, single-image defogging methods have made significant progress. Without additional information, clear scene information can be restored by relying solely on the input foggy image. At present, the mainstream defogging methods can be divided into three categories: color space-based methods, dark channel prior-based image enhancement methods, and atmospheric degradation model-based defogging methods. Generally speaking, the defogging method based on the atmospheric imaging model has better results than the first two methods, because it takes into account the degradation mechanism of haze on the image. However, all three methods have certain limitations. For example, Zhu et al. estimated the depth of field information of the foggy image by establishing a linear model, and restored the brightness of the image based on the depth of field information through the atmospheric scattering model. However, for images with slight color changes, the restored image may not eliminate color drift. Ju et al. introduced the light absorption coefficient into the atmospheric imaging model, which solved the problem of dark image restoration in traditional methods and more accurately modeled the fog degradation process. However, the restored image has exposure in areas with large scene depth. Summary of the invention

[0004] The purpose of the present invention is to provide a remote sensing image defogging method based on minimum entropy optimization to overcome the problems existing in the prior art. The present invention can optimize the scene atmospheric light value of remote sensing images in foggy weather, obtain scene atmospheric light value and transmittance that are more suitable for the scene, and effectively restore the scene information of the remote sensing image.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows: A remote sensing image defogging method based on minimum entropy optimization includes the following steps: Step 1: Obtain a foggy image under haze weather, denoted as ; Step 2: Obtain the scene atmospheric light value based on the foggy image obtained in step 1 A ; Step 3: Obtain the haze transmittance of the haze image obtained in step 1 t ; Step 4: The scene atmospheric light value obtained from step 2 A And the haze transmittance obtained in step 3 t Restore the foggy image, that is:

[0006] in, is a foggy image, is the scene atmospheric light value obtained in step 2, Restore images for foggy days, t ( i , j ) is the haze transmittance image located at ( i , j ) position.

[0007] Furthermore, in step 2, the scene atmospheric light value is obtained A , specifically: Step 2.1: Perform relative total variation smoothing on the foggy image to obtain its main structure image; Step 2.2: Perform brightness mapping on the main structure image in the horizontal and vertical directions, and use the average pixel value of each column and row of pixels in the image as the brightness mapping value of the column and row of pixels; Step 2.3: For the horizontal and vertical brightness mappings obtained in step 2.2, calculate the sum of the brightness mappings of each column and row of pixels and the brightness mappings of several adjacent columns and rows of pixels; Step 2.4: Sort the average brightness of each row and column of pixels in the image obtained in step 2.3 and the sum of the average brightness mappings of several adjacent rows and columns of pixels, find the row and column coordinates corresponding to the maximum value, and use the area adjacent to the row and column coordinates corresponding to the maximum value as the candidate rectangular area for atmospheric light. Calculate the information entropy of each row and column in the candidate rectangular area for atmospheric light, and use the number of rows and columns with the smallest information entropy as the position of the scene atmospheric light value.

[0008] Furthermore, in step 2.1, the foggy image is smoothed by relative total variation to obtain its main structure image, which is expressed as follows:

[0009] in, is the output main structure image, is the smoothing weight, To control the spatial size of the window, is the clarity parameter, is the number of iterations, RTV is the relative total variation smoothing filter; In step 2.2, the main structure image is brightness mapped in the horizontal and vertical directions, and the average pixel value of each column and row of pixels in the image is used as the brightness mapping value of the column and row of pixels, specifically:

[0010]

[0011] in, m,n Main structure images The number of rows and columns, ( i , j ) is the pixel coordinate in the main structure image, is the brightness mapping of the main structure image in the horizontal direction, is the brightness mapping of the main structure image in the horizontal direction, { r , g,b}for RGB Three color channels in color space coordinate system.

[0012] Furthermore, step 2.3 is specifically as follows: for the horizontal and vertical brightness mapping obtained in step 2.2 and , calculate the brightness map of each column and row of pixels and the adjacent The sum of the brightness maps of columns and rows of pixels:

[0013]

[0014] in, h is the neighborhood radius, row Search center for column area, value range is , col Search center for row area, value range is ; Step 2.4 is as follows: the average brightness of each row and column of pixels in the image obtained in step 2.3 and With nearby The sum of the average brightness maps of the row and column pixels and Sort and find the row and column coordinates corresponding to the maximum value. ,by Nearby The large and small areas are used as candidate rectangular areas for atmospheric light, and the information entropy of each row and column in the candidate rectangular area for atmospheric light is calculated respectively. The number of rows and columns with the smallest information entropy is used as the position of the atmospheric light value of the scene:

[0015]

[0016]

[0017]

[0018]

[0019]

[0020] In the formula, is the number of pixels of each pixel value in the image, is the total number of image pixels, P is the probability of each pixel appearing in the image, , is the information entropy of each row and column in the atmospheric light candidate rectangular area, , is the number of rows and columns with the minimum information entropy in the atmospheric light candidate rectangular area, A is the scene atmospheric light value.

[0021] Furthermore, in step 3, the haze transmittance is obtained t , specifically including: Step 3.1: The formula for defining the haze diffusion length of each pixel in the hazy image is:

[0022] in, For foggy images Medium pixel The haze diffusion length, A is the scene atmospheric light value; Step 3.2: Diffuse the haze S Transformed to the unit spherical coordinate system, 1000 sets of pitch angles and azimuth angles are uniformly sampled as the initial clustering angles in the unit spherical coordinate system, and the foggy image is clustered using the nearest neighbor search method. The pixels in are clustered into 1000 categories:

[0023]

[0024] in, is the diffusion length of haze The modulus value, and They are the pixel points in the spherical coordinate system. The elevation and azimuth angles, is the nearest neighbor search process, is the initial clustering angle, is the clustering result, k The index representing the cluster; Step 3.3: Diffuse the length of the haze S According to the clustering results Clustering is performed, and the rough estimated transmittance is calculated based on the maximum haze diffusion length in each cluster; Step 3.4: Transform the foggy image Grayscale, get its grayscale image , and Perform side window smoothing to smooth the noise while retaining its texture and depth information, and obtain the side window smoothing result image ; Step 3.5: Smooth the result image according to the side window , calculate the filter coefficient weight , and roughly estimate the transmittance Perform minimum constraint filtering to obtain the final transmittance t .

[0025] Furthermore, the rough estimated transmittance is calculated in step 3.3, specifically:

[0026]

[0027] in, k represents the index of the cluster, Cluster k The haze diffusion length of the middle pixel, for The maximum value of is a rough estimate of the transmittance.

[0028] Furthermore, step 3.4 is specifically expressed as:

[0029] The model is:

[0030] In the formula, Pixel a No. H Side windows, Pixel a Pixels in the side window b The weight of For the input image I Medium Pixels b The value of IndicatesH The smoothing result of the side window is obtained. In the calculation, the smoothed output of the side window should be consistent with the input image at the edge, so the side window output result with the smallest Euclidean distance is selected as the final output. The expression is as follows:

[0031] in, This is the result of smoothing the output of the side window.

[0032] Furthermore, step 3.5 is specifically expressed as:

[0033]

[0034] in, is the guidance coefficient, The model is:

[0035]

[0036] In the formula, and Respectively represent foggy images in haze weather and scene atmospheric light values A exist RGB The three color channel values ​​in the color space coordinate system, To roughly estimate the transmittance The result of the boundary constraints is is a smoothing term that controls the constraint strength, Pixel The index of the pixel point in a rectangular window area centered at .

[0037] Furthermore, the smoothing weight Take 0.015 to control the spatial size of the window Take 3, the number of iterations Take 4, the guide coefficient Take 0.95 to control the smoothing term of the minimum constraint filter Take 0.1.

[0038] Furthermore, the model of the relative total variation smoothing filter RTV is:

[0039] in:

[0040]

[0041]

[0042]

[0043] is the smoothing weight, is the clarity parameter, q For p The index of all pixels in a rectangular window area centered at point. Representative p A rectangular area centered on and They are x , y Partial differential in direction, g is the Gaussian kernel function:

[0044] In the formula, , , and So p , q is the coordinate value of the index, Its function is to control the spatial size of the window.

[0045] The above technical solution has the following advantages or beneficial effects: When estimating the atmospheric light value of a scene, the traditional method uses the 1% pixel values ​​with the highest brightness in the image as the atmospheric light value of the scene. This method will produce misjudgment when there are white highlight objects in the image, resulting in an overestimation of the atmospheric light value and a dark restored image. In order to solve this problem, the present invention introduces a remote sensing image defogging method based on minimum entropy optimization. First, the main structure image of the foggy image is brightness mapped in the horizontal and vertical directions, and the sum of the brightness mapping of each column and row pixel and the brightness mapping of the adjacent column and row pixels is calculated. The adjacent window area of ​​the column and row coordinates corresponding to the maximum brightness mapping value is used as the atmospheric light candidate area, and the number of rows and columns with the smallest information entropy in the candidate area is used as the location of the scene atmospheric light value. Subsequently, the present invention uses side window smoothing and minimum constraint filtering to optimize the transmittance, so that the transmittance image is smoother while retaining texture and depth information, avoiding the halo effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The figure is a flowchart of a remote sensing image dehazing method based on minimum entropy optimization.

[0047] Figure 2 and Figure 3The denoising effect of the present invention on fogged and degraded images under haze weather is compared with other image defogging methods, including: (a) original foggy image, (b) image processed by dark channel defogging method, (c) image processed by color attenuation prior method, and (d) image processed by remote sensing image defogging method based on minimum entropy optimization. DETAILED DESCRIPTION

[0048] The present invention is further described in detail below in conjunction with specific embodiments, which are intended to explain the present invention rather than to limit it. In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention. It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0049] See also Figure 1 , a remote sensing image defogging method based on minimum entropy optimization includes the following steps: Step 1: Obtain a foggy image under haze weather, denoted as ; Use image acquisition equipment to obtain foggy images in haze weather.

[0050] Step 2: Obtain the scene atmospheric light value of the foggy image obtained in step 1 A ; Step 2.1: Perform relative total variation smoothing on the foggy image obtained in step 1 to obtain its main structure image. The expression is as follows:

[0051] in, is the smoothing weight, To control the spatial size of the window, is the clarity parameter, is the number of iterations.

[0052] Step 2.2: Add the Perform brightness mapping in the horizontal and vertical directions, and use the average pixel value of each column and row of pixels in the image as the brightness mapping value of the column and row of pixels;

[0053]

[0054] in, m,n are the number of rows and columns of the image, respectively. i, j ) are pixel coordinates in the image, It is the brightness mapping of the image in the horizontal direction. is the brightness mapping of the image in the horizontal direction, for RGB Three color channels in color space coordinate system; Step 2.3: Map the horizontal and vertical brightness obtained in step 2.2 and , calculate the brightness map of each column and row of pixels and the adjacent The sum of the brightness maps of the columns and rows of pixels;

[0055]

[0056] in, h is the neighborhood radius, row Search center for column area, value range is , col Search center for row area, value range is ; Step 2.4: Calculate the average brightness of each row and column of the image obtained in step 2.3 and With nearby The sum of the average brightness maps of the row and column pixels and Sort and find the row and column coordinates corresponding to the maximum value. ,by Nearby The large and small areas are used as candidate rectangular areas for atmospheric light. The information entropy of each row and column in the candidate rectangular area for atmospheric light is calculated respectively, and the number of rows and columns with the smallest information entropy is used as the location of the atmospheric light value of the scene.

[0057]

[0058]

[0059]

[0060]

[0061]

[0062]

[0063] In the formula, ns is the number of pixels of each pixel value in the image, is the total number of image pixels, P is the probability of each pixel appearing in the image, , is the information entropy of each row and column in the atmospheric light candidate rectangular area, , is the number of rows and columns with the minimum information entropy in the atmospheric light candidate rectangular area, A is the scene atmospheric light value; Step 3: Obtain the haze transmittance of the haze image obtained in step 1 t ; Step 3.1: The formula for defining the haze diffusion length of each pixel in a hazy image is:

[0064] in, For foggy images Medium pixel The haze diffusion length, A is the scene atmospheric light value.

[0065] Step 3.2: To facilitate subsequent calculations, the haze diffusion length obtained in step 3.1 is S Transform to the unit spherical coordinate system. In the unit spherical coordinate system, 1000 sets of pitch angles and azimuth angles are uniformly sampled as the initial clustering angles, and the foggy image is clustered using the nearest neighbor search method. The pixels in are clustered into 1000 categories:

[0066]

[0067] in, is the diffusion length of haze The modulus value, and They are the pixel points in the spherical coordinate system. The elevation and azimuth angles. is the nearest neighbor search process, is the initial clustering angle, is the clustering result, k The index representing the cluster; Step 3.3: Substitute the haze diffusion length obtained in step 3.1 S , according to the clustering results obtained in step 3.2 Clustering is performed, and the rough estimated transmittance is calculated based on the maximum haze diffusion length in each cluster. The expression is as follows:

[0068]

[0069] in, k represents the index of the cluster, Cluster k The haze diffusion length of the middle pixel, for The maximum value of is a rough estimate of the transmittance.

[0070] Step 3.4: Transform the foggy image Grayscale, get its grayscale image , and Perform side window smoothing to smooth out noise while preserving its texture and depth information:

[0071] The model is:

[0072]

[0073]

[0074] In the formula, Pixel a No. H Side windows, Pixel a Pixels in the side window b The weight of For the input image I Medium Pixels b The value of Indicates H In the calculation, the output of the side window smoothing should be consistent with the input image at the edge, so the side window output with the smallest Euclidean distance is selected as the final output. The expression is as follows:

[0075] in, This is the result of smoothing the output of the side window.

[0076] Step 3.5: The side window smoothing result obtained in step 3.4 , calculate the filter coefficient weight , and the rough estimated transmittance obtained in step 3.3 Perform minimum constraint filtering to obtain the final transmittance t .

[0077]

[0078]

[0079] in, is the guidance coefficient, The model is:

[0080]

[0081] In the formula, and Respectively represent the foggy images obtained in step 1 and 2 under haze weather and scene atmospheric light values A exist RGB The three color channel values ​​in the color space coordinate system. To roughly estimate the transmittance The result of the boundary constraints is is a smoothing term that controls the constraint strength, Pixel The index of the pixel point in a rectangular window area centered at .

[0082] Step 4: The scene atmospheric light value obtained from step 2 A And the final transmittance obtained in step 3 t Restore the foggy image, that is:

[0083] in, is a foggy image, is the scene atmospheric light value obtained in step 2, Restore images for foggy days.

[0084] Figure 2 The processing effect is intuitively demonstrated. Figure 2 (a) is the original image, in which there is a white highlight area in the foreground. The result obtained by processing the image using the dark channel dehazing method is as follows: Figure 2As shown in (b), due to the presence of white highlights in the foreground of the image, the atmospheric light value of the selected scene is too large, the restoration effect of the entire image is dark, and the colors of the sky and the foreground ground are distorted. The result after using the dehazing method with color attenuation prior is shown in Figure 2 As shown in (c), it can be observed that the defogging effect of this method on the distant mountain peaks is not thorough and color shift occurs on the near ground. The results obtained by using the defogging method of the present invention are as follows Figure 2 As shown in (d), compared with the traditional method, it is more accurate in selecting the atmospheric light value, and while ensuring the defogging effect in the distant view, it also effectively avoids the color shift phenomenon in the near view.

[0085] Table 1 Evaluation coefficients of comparison group 2

[0086] This table compares the effects of processed images by comparing grayscale image contrast, average gradient and haze concentration evaluator. It can be seen from this table that the average gradient and grayscale image contrast of the original image are low, and the FADE of the haze concentration evaluator is relatively high. After processing by the dark channel, color attenuation prior and this method, the average gradient and grayscale image contrast are significantly improved, and FADE is significantly reduced.

[0087] Figure 3 The processing effect is intuitively demonstrated. Figure 3 (a) is the original image. The result obtained by processing the image using the simple dark channel method is as follows Figure 3 As shown in (b), the atmospheric light value is selected without considering the local information of the image, resulting in the atmospheric light value being too large, and the restoration effect of the entire image is dark. The result after using the dehazing method with color attenuation prior is shown in Figure 3 As shown in (c), it can be observed that this method does not completely remove fog in the distant view. The results obtained by using the defogging method of the present invention are as follows Figure 3 As shown in (d), compared with traditional methods, it is more accurate in selecting atmospheric light values, and it highlights the texture details of the image while ensuring the defogging effect in the distance.

[0088] Table 2 Evaluation coefficients of comparison group 2

[0089] It can be seen from this table that the average gradient and grayscale image contrast of the original image are low, and the FADE of the haze concentration estimator is relatively high. After being processed by the dark channel, color attenuation prior and this method, the average gradient and grayscale image contrast are significantly improved, and the FADE is significantly reduced.

[0090] In summary, this method can accurately select the scene atmospheric light value of foggy images and achieve ideal defogging effect, which is of great significance for clarifying foggy remote sensing images and obtaining remote sensing image information.

[0091] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A remote sensing image defogging method based on minimum entropy optimization, characterized in that: The following steps are involved: Step 1: Obtain a foggy image under haze weather, denoted as ; Step 2: Obtain the scene atmospheric light value based on the foggy image obtained in step 1 A ; Step 3: Obtain the haze transmittance of the haze image obtained in step 1 t ; Step 4: The scene atmospheric light value obtained from step 2 A And the haze transmittance obtained in step 3 t Restore the foggy image, that is: in, is a foggy image, is the scene atmospheric light value obtained in step 2, Restore images for foggy days, t ( i , j ) is the haze transmittance image located at ( i , j ) position.

2. The remote sensing image defogging method based on minimum entropy optimization according to claim 1, characterized in that: Get the scene atmospheric light value in step 2 A , specifically: Step 2.1: Perform relative total variation smoothing on the foggy image to obtain its main structure image; Step 2.2: Perform brightness mapping on the main structure image in the horizontal and vertical directions, and use the average pixel value of each column and row of pixels in the image as the brightness mapping value of the column and row of pixels; Step 2.3: For the horizontal and vertical brightness mappings obtained in step 2.2, calculate the sum of the brightness mappings of each column and row of pixels and the brightness mappings of several adjacent columns and rows of pixels; Step 2.4: Sort the average brightness of each row and column of pixels in the image obtained in step 2.3 and the sum of the average brightness mappings of several adjacent rows and columns of pixels, find the row and column coordinates corresponding to the maximum value, and use the area adjacent to the row and column coordinates corresponding to the maximum value as the candidate rectangular area for atmospheric light. Calculate the information entropy of each row and column in the candidate rectangular area for atmospheric light, and use the number of rows and columns with the smallest information entropy as the position of the scene atmospheric light value.

3. The remote sensing image defogging method based on minimum entropy optimization according to claim 2, characterized in that: In step 2.1, the foggy image is smoothed by relative total variation to obtain its main structure image, and the expression is as follows: in, is the output main structure image, is the smoothing weight, To control the spatial size of the window, is the clarity parameter, is the number of iterations, RTV is the relative total variation smoothing filter; In step 2.2, the main structure image is brightness mapped in the horizontal and vertical directions, and the average pixel value of each column and row of pixels in the image is used as the brightness mapping value of the column and row of pixels, specifically: in, m,n Main structure images The number of rows and columns, ( i , j ) is the pixel coordinate in the main structure image, is the brightness mapping of the main structure image in the horizontal direction, is the brightness mapping of the main structure image in the horizontal direction, { r , g,b }for RGB Three color channels in color space coordinate system.

4. The remote sensing image defogging method based on minimum entropy optimization according to claim 3, characterized in that: Step 2.3 is as follows: for the horizontal and vertical brightness mapping obtained in step 2.2 and , calculate the brightness map of each column and row of pixels and the adjacent The sum of the brightness maps of columns and rows of pixels: in, h is the neighborhood radius, row Search center for column area, value range is , col Search center for row area, value range is ; Step 2.4 is as follows: the average brightness of each row and column of pixels in the image obtained in step 2.3 and With nearby The sum of the average brightness maps of the row and column pixels and Sort and find the row and column coordinates corresponding to the maximum value. ,by Nearby The large and small areas are used as candidate rectangular areas for atmospheric light, and the information entropy of each row and column in the candidate rectangular area for atmospheric light is calculated respectively. The number of rows and columns with the smallest information entropy is used as the position of the atmospheric light value of the scene: In the formula, is the number of pixels of each pixel value in the image, is the total number of image pixels, P is the probability of each pixel appearing in the image, , is the information entropy of each row and column in the atmospheric light candidate rectangular area, , is the number of rows and columns with the minimum information entropy in the atmospheric light candidate rectangular area, A is the scene atmospheric light value.

5. The remote sensing image defogging method based on minimum entropy optimization according to claim 2, characterized in that: Obtain haze transmittance in step 3 t , specifically including: Step 3.1: The formula for defining the haze diffusion length of each pixel in the hazy image is: in, For foggy images Medium pixel The haze diffusion length, A is the scene atmospheric light value; Step 3.2: Diffuse the haze S Transformed to the unit spherical coordinate system, 1000 sets of pitch angles and azimuth angles are uniformly sampled as the initial clustering angles in the unit spherical coordinate system, and the foggy image is clustered using the nearest neighbor search method. The pixels in are clustered into 1000 categories: in, is the diffusion length of haze The modulus value, and They are the pixel points in the spherical coordinate system. The elevation and azimuth angles, is the nearest neighbor search process, is the initial clustering angle, is the clustering result, k The index representing the cluster; Step 3.3: Diffuse the length of the haze S According to the clustering results Clustering is performed, and the rough estimated transmittance is calculated based on the maximum haze diffusion length in each cluster; Step 3.4: Transform the foggy image Grayscale, get its grayscale image , and Perform side window smoothing to smooth the noise while retaining its texture and depth information, and obtain the side window smoothing result image ; Step 3.5: Smooth the result image according to the side window , calculate the filter coefficient weight , and roughly estimate the transmittance Perform minimum constraint filtering to obtain the final transmittance t .

6. The remote sensing image defogging method based on minimum entropy optimization according to claim 5, characterized in that: The rough estimated transmittance is calculated in step 3.3, specifically: in, k represents the index of the cluster, Cluster k The haze diffusion length of the middle pixel, for The maximum value of is a rough estimate of the transmittance.

7. The remote sensing image defogging method based on minimum entropy optimization according to claim 5, characterized in that: Step 3.4 is specifically expressed as: The model is: In the formula, Pixel a No. H Side windows, Pixel a Pixels in the side window b The weight of For the input image I Medium Pixels b The value of Indicates H The smoothing result of the side window is obtained. In the calculation, the smoothed output of the side window should be consistent with the input image at the edge, so the side window output result with the smallest Euclidean distance is selected as the final output. The expression is as follows: in, This is the result of smoothing the output of the side window.

8. The remote sensing image defogging method based on minimum entropy optimization according to claim 5, characterized in that: Step 3.5 is specifically expressed as: in, is the guidance coefficient, The model is: In the formula, and Respectively represent foggy images in haze weather and scene atmospheric light values A exist RGB The three color channel values ​​in the color space coordinate system, To roughly estimate the transmittance The result of the boundary constraints is is a smoothing term that controls the constraint strength, Pixel The index of the pixel point in a rectangular window area centered at .

9. The remote sensing image defogging method based on minimum entropy optimization according to claim 8, characterized in that: The smoothing weight Take 0.015 to control the spatial size of the window Take 3, the number of iterations Take 4, the guide coefficient Take 0.95 to control the smoothing term of the minimum constraint filter Take 0.

1.

10. The remote sensing image defogging method based on minimum entropy optimization according to claim 3, characterized in that: The model of the relative total variation smoothing filter RTV is: in: is the smoothing weight, is the clarity parameter, q For p The index of all pixels in a rectangular window area centered at point. Representative p A rectangular area centered on and They are x , y Partial differential in direction, g is the Gaussian kernel function: In the formula, , , and So p , q is the coordinate value of the index, Its function is to control the spatial size of the window.

Images