Method for clearing dust and fog image of coal mining face based on atmospheric scattering model

Abstract

The application discloses a dust-mist image clearification method for a coal mining face based on an atmospheric scattering model, which comprises the following steps: according to image channel differences and brightness information, a dust-mist area in a dust-mist image of the coal mining face is divided into a thick fog area and a non-thick fog area; an initial illumination map of the dust-mist image of the coal mining face is estimated by using a Max- RGB method, and the initial illumination map is finely processed to obtain a global illumination map; a transmittance value of the thick fog area is estimated by using an optimized color attenuation model; a transmittance value of the non-thick fog area is calculated by using a dark channel priori and an ambient light of the area; ambient light values of the thick fog area and the non-thick fog area are respectively calculated by using the global illumination map; ambient light and transmittance of different areas are subjected to Alpha fusion; and the global ambient light value and the transmittance value are substituted into an atmospheric scattering model to restore a low-illumination dust-mist image. The application effectively solves the problem of difficult parameter estimation of the atmospheric scattering model.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of coal mine underground image sharpening, and particularly relates to a dust-mist image sharpening method for a coal mining face based on an atmospheric scattering model. BACKGROUND

[0002] At present, the research on the single image defogging and dust removing method mainly falls into two directions: one is the defogging method based on image enhancement technology; and the other is the defogging method based on image restoration technology. The defogging method based on enhancement technology develops the fastest, and its defogging idea is very intuitive, that is, the influence of the interference information in the image is eliminated, the contrast, brightness and other information of the image is restored, and then the overall visual effect of the image is improved. The classical methods include the histogram equalization method for defogging according to the distribution characteristics of the image histogram, the Retinex method for defogging by using the color constancy theory, the homomorphic filtering method for defogging by combining the image gray variation and frequency information, and the wavelet transform method for defogging according to the spatial distribution characteristics and frequency domain characteristics of the image, etc. However, this kind of method does not establish a fogging model in a real scene, ignores the mechanism of image degradation under the condition of dust and mist, and cannot realize the image defogging in a true physical sense. The image restoration method based on the imaging physical model under the condition of dust and mist analyzes the causes of image degradation by using the atmospheric scattering model, so as to realize the image defogging in a true sense.

[0003] Therefore, the defogging method based on the image restoration technology has a clear mathematical model, and starts from the essence of the dust-mist image, so that the visual effect of the final defogging image is more natural, and the defogging effect is more obvious. Since the atmospheric scattering model is completely derived from the physical model, it has inherent superiority compared with the image enhancement method. At present, the defogging methods based on image restoration mainly include: (1) the defogging method based on solving the partial differential equation, the corresponding energy model of the overall or local characteristics of the image is established by extracting the feature information and illumination information of the image, so as to obtain the partial differential equation reflecting the internal characteristics of the image, and the image scene depth or gradient and other parameters are obtained by solving the equation, so as to complete the task of image defogging; (2) the defogging method based on depth estimation, the depth information of the image is obtained by the image depth estimation method, so as to estimate the related parameters of the atmospheric scattering model, and obtain the defogged image; (3) the defogging method based on prior information, the parameters in the atmospheric scattering model are further estimated by using some prior knowledge such as the characteristics of the foggy image, and the fog-free image is restored.

[0004] Because the brightness of the downhole environment is lower, the fog is thicker, and the illumination is more uneven, the difficulty of defogging is also greatly improved. At present, the methods for clearification of downhole dust fog images mainly include: (1) a clearification method based on atmospheric scattering model and principal component analysis, based on atmospheric scattering model and dark channel prior, the image information is weighted processed by principal component analysis method, so as to estimate the atmospheric light value, and realize the clearification of the mine downhole image. (2) a downhole fog image clearification method based on contrast enhancement, by improving the contrast, brightness and other characteristics of the dust fog image, the information entropy of the restored image is improved to a certain extent. (3) an image clearification method based on atmospheric scattering model and total variation regularization, the initial transmittance matrix is smoothed by using total variation regularization, and the fine transmittance is obtained, and then the clear image is restored according to the atmospheric scattering model.

[0005] Due to the mining activities of the coal mining face, the monitoring video image will produce low illumination, uneven illumination, high dust fog and uneven distribution, which is not conducive to the safety production of coal mine, and it is difficult to realize the remote control of equipment operators, the use of intelligent video recognition technology, and the like, and the image clearification of the coal mining face is particularly important. However, the coal mining face has a complex environment, the illumination in the operation area is uneven, the dust fog is dense and the medium is unevenly distributed, therefore, the effect of the current method for clearification of the dust fog image of the coal mining face is not ideal, and the estimation of the ambient light value and the transmittance parameter in the atmospheric scattering model becomes a difficult problem. SUMMARY

[0006] The purpose of the present application is to provide a dust fog image clearification method for a coal mining face based on an atmospheric scattering model, which divides the dust fog image into a thick fog region and a non-thick fog region according to the brightness and channel difference of the downhole dust fog image, combines the image features of different regions, respectively calculates the atmospheric scattering model parameters of each region, and fuses the global model parameters to solve the technical problem of inaccurate model parameter estimation in a complex coal mining environment.

[0007] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0008] The dust fog image clearification method for a coal mining face based on an atmospheric scattering model comprises the following steps:

[0009] Step 1. By analyzing the distribution characteristics of dust and water mist in the coal mining face, the dust fog region in the coal mining face dust fog image is divided into a thick fog region and a non-thick fog region according to the image channel difference and brightness information;

[0010] Step 2. The Max-RGB method is used to estimate the initial illumination map of the coal mining face dust fog image, and the initial illumination map is finely processed to obtain a fine illumination map, i.e. a global illumination map;

[0011] Step 3. In the thick fog area, the transmittance value of the thick fog area is estimated by using an optimized color attenuation model; in the non-thick fog area, the transmittance value of the non-thick fog area is calculated by using the dark channel prior and the ambient light of the area;

[0012] Step 4. In the thick fog area, the ambient light value of the thick fog area is calculated by using the global illumination map and the transmittance of the thick fog area; in the non-thick fog area, the ambient light value of the non-thick fog area is obtained by using the global illumination map;

[0013] Step 5. The ambient light and the transmittance of different areas are Alpha fused, and the noise generated in the fusion process is suppressed while the edge information of the image is retained by using guided filtering, to obtain the global ambient light value and the transmittance value;

[0014] The global ambient light value and the transmittance value are substituted into the atmospheric scattering model to restore the low-illumination dust fog image.

[0015] The present application has the following advantages:

[0016] As described above, the present application proposes a dust fog image clarification method for a coal mining face based on an atmospheric scattering model, aiming at the defogging task of the image of the coal mining face in a coal mine. Since a large amount of dust and water mist is generated in the operation process, the dust and water mist particles are densely distributed in the area, and the dust and water mist thickness is uneven. In order to better restore the image of the thick fog area, the dust fog area in the dust fog image of the coal mining face is first segmented into a thick fog area and a non-thick fog area according to the channel difference and the brightness information of the image. The initial illumination map of the dust fog image of the coal mining face is estimated by using the Max-RGB method, and the initial illumination map is finely processed to obtain a global illumination map. In the thick fog area, the transmittance value of the thick fog area is estimated by using an optimized color attenuation model; in the non-thick fog area, the transmittance value of the non-thick fog area is calculated by using the dark channel prior and the ambient light of the area. The ambient light values of the thick fog area and the non-thick fog area are calculated by using the global illumination map. The ambient light and the transmittance of different areas are Alpha fused. The global ambient light value and the transmittance value are substituted into the atmospheric scattering model to restore the low-illumination dust fog image. Finally, the restored image obtained by the method of the present application is compared with the restored image obtained by other methods, and it can be seen that the method of the present application can effectively suppress the dust fog concentration of the restored image, and has strong image illumination restoration ability and thick fog area image detail information restoration ability. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The flowchart of the dust fog image clarification method for a coal mining face based on an atmospheric scattering model in the embodiment of the present application is shown in the figure;

[0018] Figure 2 The illumination map estimation flowchart of the underground dust fog image in the embodiment of the present application is shown in the figure;

[0019] Figure 3 The figure is a part of experimental data set in the present application;

[0020] Figure 4 The figure is a comparison of the defogging effect of the method of the present application and other two existing methods. DETAILED DESCRIPTION

[0021] Noun explanation:

[0022] The dust and fog concentration distribution of the image of the coal mining face is obviously uneven. A large amount of dust and fog is generated near the coal mining operation area, and the density is uneven, which is defined as a dense fog area. Away from the area, the dust and fog is relatively thin and evenly distributed, which is defined as a non-dense fog area.

[0023] The present embodiment describes a dust and fog image clarification method for a coal mining face based on an atmospheric scattering model, to solve the problem of inaccurate estimation of environmental light value and transmittance in the atmospheric scattering model under complex coal mining environment.

[0024] The method generally includes the following three parts:

[0025] The first is to segment the image area according to the dust and fog concentration of the coal mining image, the second is to estimate the atmospheric light value and transmittance of each area, and the third is to restore a clear image based on the atmospheric scattering model after fusing the area parameters.

[0026] The present application will be further described in detail in combination with the drawings and specific embodiments:

[0027] As shown in Figure 1 The dust and fog image clarification method for a coal mining face based on an atmospheric scattering model includes the following steps:

[0028] Step 1. By analyzing the distribution characteristics of dust and fog in the coal mining face, according to the channel difference and brightness information of the image, the dust and fog area in the dust and fog image of the coal mining face is segmented into a dense fog area and a non-dense fog area.

[0029] In the dust and fog image, the channel difference is used to represent the difference between the maximum color channel value and the minimum color channel value of each area of the image. The dust and fog particles in the dense fog area are densely distributed, the scattered light intensity is high, the overall brightness value of the image is large, and the channel difference value is small. According to the above characteristics, the dust and fog image area is segmented. The image channel difference is defined as:

[0030]

[0031] Where C d (x,y) is the image channel difference, C represents the color channel, I C (x,y) represents the input image channel map, The maximum color channel value of each region of the image is obtained, The minimum color channel value of each region of the image is obtained.

[0032] The channel difference of the dense fog region can be approximated as 0. However, the channel difference of part of the non-dense fog region, such as the coal wall background region, is also small, and therefore, the brightness feature needs to be used for further segmentation.

[0033] The present application uses gamma transformation to improve the brightness difference of the image, and the brightness of the transformed image is represented as:

[0034]

[0035] wherein η(x,y) is the brightness map after gamma transformation, and η(x,y) is normalized as shown in formula (3):

[0036]

[0037] wherein, is the normalized brightness map, and η min represents the minimum value in the brightness map, and η max represents the maximum value in the brightness map, the overall brightness of the dense fog region image is high, and the channel difference C d (x,y)→0.

[0038] Based on the above priori, the image dust fog region probability distribution function is constructed as shown in formula (4):

[0039]

[0040] In order to improve the signal-to-noise ratio of each region of the segmented image, the maximum entropy threshold segmentation method is used to determine the threshold, and the probability distribution function of the whole dust fog image is segmented to obtain the dense fog region and the non-dense fog region.

[0041] At the same time, α(x,y) is used as a weighting coefficient in the following Alpha fusion process for different regions.

[0042] Step 2. The initial illumination map of the coal mining face dust fog image is estimated by using the Max-RGB method. In order to better preserve the structure information and edge information of the dust fog image, the initial illumination map is refined to obtain a refined illumination map.

[0043] The light intensity of the imaging region is determined by the ambient light in the imaging process.

[0044] The ambient light value at each position can be inversely solved by estimating the global illumination map. The underground dust fog image has the problems of low illumination and uneven illumination, and the existing illumination map estimation method does not have ideal processing effect for the underground image.

[0045] To solve the above problems, the application provides a downhole dust and fog image illumination map estimation method, and a method flow is as shown in Figure 2

[0046] Step 2.1. Setting an initial illumination map.

[0047] In the field of image restoration, the Max-RGB method is often used to estimate the illumination map of an image, which takes the maximum value of the R, G and B color channels as the illumination value, but the illumination map estimated by this method is only suitable for scenes with large illumination.

[0048] For solving the illumination map of a downhole low-illumination image, the illumination map obtained by the method needs to be refined.

[0049] First, the Max-RGB method is used to estimate the initial illumination map of the dust and fog image of the coal mining face.

[0050] Let the initial illumination map P0(x,y) be:

[0051]

[0052] Step 2.2. Solving the refined illumination map. The initialization illumination map does not consider the image structure information and cannot preserve the image edges. In order to solve this problem, the application constructs an optimization objective function that preserves the overall structure and edge information of the image as follows:

[0053]

[0054] Wherein, represents the Frobenius norm, indicates the image structure preservation term, W d (x,y) indicates the weight matrix, P(x,y) indicates the refined illumination map, and ||W d (x,y)P(x,y)||1 indicates the l1 norm.

[0055] In order to preserve the overall structure of the illumination map, the initial illumination map obtained in step 2.1 and the refined illumination map P(x,y) to be solved are combined to obtain At the same time, in order to preserve the edge information of the image in the illumination map, the weight matrix W d (x,y) is constructed, and the weight matrix is combined with the refined illumination map P(x,y) to obtain ||W d (x,y)P(x,y)||1.

[0056] The setting of the weight matrix W d (x,y) is crucial for the edge gradient recovery of the illumination map, and the weight matrix W d (x,y) is based on the relative total variation model, and the weight matrix W d ​The expression of (x, y) is:

[0057]

[0058] Where, Ω(x, y) represents a neighborhood centered at (x, y), G σ (x, y) is a Gaussian kernel function with a standard deviation of σ, is a gradient sign, d represents a gradient direction, P0(z) represents an initial light map, and z is a coordinate in Ω(x, y).

[0059] G σ (x, y) is expressed as:

[0060]

[0061] ||W d An approximation of (x, y)P(x, y)||1 is expressed as:

[0062]

[0063] Where, h and v represent horizontal and vertical directions respectively, and in order to avoid a zero denominator in calculation, a number close to zero, ∈, ∈>0, needs to be added to the denominator; and the optimization objective function is rewritten as:

[0064]

[0065] λ is a scale coefficient used to balance the two penalty factors before and after. In order to solve formula (10), according to the Lagrange theorem, we have:

[0066]

[0067] Let Write formula (11) as a discrete form, and calculate the approximate solution of the above equation as:

[0068]

[0069] Where, D + represents a forward finite difference, D - represents a backward finite difference, Δt is an iteration step length, P n (x, y) and P n+1 (x, y) are the refined light maps of the n-th estimation and the n+1-th estimation respectively; it is assumed that formula (6) reaches a minimum value or reaches a maximum iteration number at the N-th iteration, and P n (x, y) is the refined light map P(x, y) to be solved.

[0070] The underground dust fog image illumination map estimation method provided by step 2 retains the overall structure of the image and effectively improves the brightness of the edge region of the image compared with the traditional illumination map estimation method.

[0071] Step 3. In the thick fog area, the transmittance value of the thick fog area is estimated by using the optimized color attenuation model; in the non-thick fog area, the transmittance value of the non-thick fog area is calculated by using the dark channel prior and the ambient light matrix of the area.

[0072] Step 3.1. Estimation of the transmittance value of the thick fog area.

[0073] The transmittance in the atmospheric attenuation model is related to the scene depth and the atmospheric scattering coefficient, and is expressed as:

[0074] t(x,y)=e -βd(x,y) (13)

[0075] Wherein, β is the atmospheric scattering coefficient, and d(x,y) is the scene depth. The brightness, saturation and gradient of the image in the thick fog area of the coal mining face approximately conform to the color attenuation model, and the scene depth in the area is linearly represented by the gradient, brightness and saturation.

[0076] According to the linear image depth model of the color attenuation model, the scene depth in the thick fog area is expressed as:

[0077] d(x,y)=ω1L(x,y)+ω2S(x,y)+ω3G(x,y)+ε,(x,y)∈DFR (14)

[0078] Wherein, ω1, ω2 and ω3 are linear coefficients, ε represents an estimation error, DFR represents the thick fog area, and d(x,y), L(x,y), S(x,y) and G(x,y) represent the depth, brightness, saturation and gradient information of the image respectively.

[0079] A fixed-size window range is selected to estimate the average depth of the dust fog :

[0080]

[0081] Wherein, Φ i (x,y) is a window centered at (x,y), N is the number of windows, represents the average depth of the window, L(x,y), S(x,y) and G(x,y) are the average brightness, average saturation and average gradient of the window respectively.

[0082] The physical meaning of the atmospheric scattering coefficient β is the scattering ability of the suspended particles in a unit volume to the surrounding illumination. The traditional color attenuation model considers that the dust fog medium is uniformly distributed in the imaging area, that is, β is a constant value.

[0083] The dust-mist density in the thick fog area is uneven, and the atmospheric scattering coefficient is related to the dust-mist density. The higher the dust-mist density in the thick fog area is, the greater the dust-mist image brightness is. Therefore, the atmospheric scattering coefficient β(x, y) in the embodiment is represented as:

[0084] β(x, y) = a·e b·L(x,y) (16)

[0085] Wherein, a and b are coefficients, and L(x, y) represents the brightness image.

[0086] The formula assumes that the atmospheric scattering coefficient is proportional to the fog density, and references the relationship between the scene depth and the fog density to obtain the formula. The effectiveness of the formula is proved in combination with the experimental results. After the scene depth and the atmospheric scattering coefficient are respectively obtained according to the formulas (14) and (16), the transmittance t0(x, y) of the thick fog area is obtained by substituting the transmittance matrix into the formula (15).

[0087] The transmittance matrix is guided and filtered to retain the position detail information, and the brightness, gradient and other information are recovered in combination with the features of adjacent pixel points. The thick fog suppression effect of the embodiment is good, and the gradient, saturation and other information of the thick fog area can be effectively recovered.

[0088] Step 3.2. Estimation of the transmittance value of the thick fog area.

[0089] In the non-thick fog area, the saturation, brightness and other features of the image do not meet the applicable conditions of the color decay model, but meet the dark channel prior condition. For the non-thick fog area, the dark channel prior is used to obtain the transmittance t1(x, y) of the non-thick fog area as:

[0090]

[0091] Wherein, Ω(x, y) represents a window centered on the pixel (x, y), represents each channel ambient light map.

[0092] Step 4. In the thick fog area, the ambient light value of the thick fog area is calculated by using the global light map and the transmittance of the thick fog area; in the non-thick fog area, the ambient light value of the non-thick fog area is obtained by using the global light map.

[0093] The step 4 is specifically:

[0094] Step 4.1. Ambient light estimation of the thick fog area.

[0095] In the outdoor atmospheric environment, the ambient light value is often taken as a global constant value. However, the coal mining face underground adopts artificial light source, and there is a problem of uneven illumination, so the ambient light value A needs to be expanded into an ambient light matrix A(x, y).

[0096] The light intensity in the thick fog area is higher, and the dust and fog particles are densely distributed. The light mainly comes from the ambient light which attenuates with the scene depth. The transmittance in the atmospheric scattering model represents the attenuation process of the ambient light. The light distribution in the thick fog area is expressed as:

[0097] P(x, y) = A0(x, y) t0(x, y), (x, y) e DFR (18)

[0098] where A0(x, y) represents the ambient light value in the thick fog area, t0(x, y) represents the transmittance in the thick fog area, and DFR represents the thick fog area. After the global light map P(x, y) is estimated in step 2, the ambient light value A0(x, y) in the thick fog area is:

[0099] A0(x, y) = P(x, y) / t0(x, y) (19)

[0100] The transmittance t0(x, y) in the thick fog area is obtained by step 3.1, and then the ambient light value in the thick fog area is obtained, so that the clear processing of the thick fog area is realized.

[0101] Step 4.2. Ambient light estimation in non-thick fog area.

[0102] Compared with the thick fog area, the ambient light intensity in the non-thick fog area is smaller, and the ambient light mainly depends on the scattering effect. The ambient light value A1(x, y) in the non-thick fog area is:

[0103] Step 5. Alpha blending of the ambient light matrix and the transmittance matrix in different areas, and using guided filtering to suppress the noise generated in the fusion process while preserving the edge information of the image, to obtain the global ambient light value and the transmittance value;

[0104] Substitute the global ambient light value and the transmittance value into the atmospheric scattering model to restore the low-illumination dust and fog image.

[0105] The step 5 is specifically:

[0106] In the calculation of the global ambient light and the transmittance matrix, in order to suppress the edge effect, the ambient light value and the transmittance in different areas are Alpha blended respectively by using the thick fog area probability function a(x, y) to obtain the global ambient light value A(x, y) and the transmittance t(x, y). The calculation formula is shown in formula (21).

[0107]

[0108] The global transmittance is guided filtered to reduce the noise generated in the fusion process while keeping the edge. The low-illumination dust and fog image is restored by using the atmospheric scattering model, as shown in formula (22).

[0109]

[0110] wherein, J(x, y) represents the restored sharpening image; max(t(x, y), η) represents the maximum value of t(x, y) and η.

[0111] η represents a transmittance correction parameter, in order to avoid that the transmittance value is too small to cause the restored image to be overexposed, the transmittance correction parameter η is for example 0.1, so as to ensure that the minimum value of the transmittance should not be lower than 0.1.

[0112] In addition, in order to verify the dust and mist removal effect of the method of the present application on the underground image, the present application selects the image containing dust and mist under the mine, and the experimental data set is derived from the underground operation video of XXX coal mine. The data set contains 60 images, which are divided into three categories of coal mining machine, head, and tunnel according to different operation positions, and part of the data set is shown in Table 1.

[0113] The computer used in the experiment is configured as CPU Intel(R) Core(TM) i5-8250U 1.80GHz, RAM 8GB. The experimental environment is Ubuntu 16.04LTS, and the software platform is JetBrains CLion 2020.3.

[0114] The experimental data set is derived from the dust and mist images at different positions of the coal mine underground coal mining face.

[0115] The data set includes 1000 underground dust and mist images, which are mainly divided into four categories:

[0116] The first category is dust and mist image with uneven illumination; the second category is high dust image; the third category is water mist image generated by dust removal device; and the fourth category is uneven dust and mist image. Part of the experimental data set is shown in Table 2. Figure 3 For the above data set, the dehazing results of the underground dust and mist image are obtained through the experiment, and are compared with other widely used dehazing methods such as the dehazing method based on traditional color attenuation prior (Zhu method) and the night image dehazing method based on maximum reflection prior (Jing method).

[0117] The above two methods are both image dehazing methods based on atmospheric scattering model, wherein the Zhu method is verified to be better in processing thick dust and mist area, and the Jing method is better for night mist image dehazing.

[0118] Four images in different positions in the data set are selected for subjective evaluation of the dehazing method, and the experimental results are shown in Table 3. Figure 4 Figure 4 ​As can be seen, for the defogging work of such special environment of underground coal mine, the traditional defogging method, such as Zhu method, although can achieve the effect of defogging in some areas, the overall brightness of the image is low, the information recovery ability of the image after defogging is insufficient, the defogging effect of the thick fog area is not obvious, and the defogging requirement is not completed. After processing by Jing method, the overall brightness of the image is enhanced, but the processed image has color deviation problem (for example, the conveying belt area in the second type of image, the tarpaulin area in the third type of image), the transition of different areas is uneven, the color of the image after defogging is not natural, and the texture details of the image are not reserved (for example, the coal block part of the first type of image), and the processing effect of Jing method on the thick fog area is also poor (for example, the coal wall part of the fourth type of image), and there is a serious dust and fog residual phenomenon.

[0119] Through experimental verification, the method of the present application can preserve the overall structure information and edge information of the image after processing the underground dust and fog image, and there is no color deviation problem, the overall brightness of the image is moderate, the gradient information of the thick fog area can be well recovered, and the defogging requirement of the underground coal mine is completed.

[0120] In order to further prove the effect of the defogging method of the present application, three kinds of no-reference objective evaluation indexes are used to evaluate the method, as shown in Table 1, which are new visible edge number, fog density estimation (FADE) and average saturation.

[0121] The new visible edge number describes the edge preservation performance and detail recovery ability of the method, and the larger the value is, the stronger the ability of the method to preserve the edge information is; the fog density estimation estimates the fog density of the input image by calculating the fog perception feature of the input image and comparing the deviation of the fog perception feature extracted from the input image and the no-fog image, and the smaller the value is, the better the defogging effect is; the average saturation describes the color recovery ability of the defogged image, and the larger the value is, the more colorful the color is.

[0122] Table 1 Objective evaluation of image defogging method

[0123]

[0124] As can be seen from the three kinds of objective evaluation indexes, for the underground dust and fog image, the three methods can effectively reduce the dust and fog concentration, the average saturation of the images processed by Jing method and Zhu method is slightly improved, but the overall structure and edge information of the image cannot be preserved, and the recovery ability of the image detail information is not strong; the overall brightness of the image after defogging by Zhu method is low, resulting in a large amount of information lost in the defogged image. The dust and fog concentration suppression effect of the method of the present application is better than that of the other two methods, the brightness of the recovered image is relatively high, and the edge recovery ability of the method of the present application is stronger, which can effectively extract the image gradient information and other information of the uneven thick fog area. In summary, the performance of the method of the present application is more excellent, and the image restoration effect is more prominent.

[0125] Of course, the above description is merely preferred embodiments of the present application, and the present application is not limited to the above-described embodiments. It should be understood that any modifications, equivalent replacements, and obvious variations made by those skilled in the art based on the teachings of the present specification fall within the scope of the present application, and should be protected by the present application.

Claims

1. A method for sharpening dust and fog images in coal mining faces based on an atmospheric scattering model, characterized in that, Includes the following steps: Step 1. By analyzing the distribution characteristics of dust and water mist at the coal mining face, and based on the differences in image channels and brightness information, the dust and fog areas in the dust and fog images of the coal mining face are divided into dense fog areas and non-dense fog areas; Step 2. The initial illumination map of the dust and fog image of the coal mining face is estimated using the Max-RGB method, and the initial illumination map is refined to obtain the refined illumination map, i.e., the global illumination map. Step 3. In the dense fog area, the transmittance value of the dense fog area is estimated using an optimized color attenuation model; in the non-dense fog area, the transmittance value of the non-dense fog area is calculated using the dark channel prior and the ambient light of the area. Step 4. In dense fog areas, calculate the ambient light value of the dense fog area using the global illumination map and the transmittance of the dense fog area; in non-dense fog areas, obtain the ambient light value of the non-dense fog area using the global illumination map. Step 5. Perform Alpha fusion on the ambient light and transmittance of different regions, and use guided filtering to suppress the noise generated during the fusion process while preserving the image edge information, to obtain the global ambient light value and transmittance value. The global ambient light and transmittance values ​​are substituted into the atmospheric scattering model to restore the sharpened image; Step 1 specifically involves: In dust and fog images, channel difference is used to represent the difference between the maximum and minimum color channel values ​​in each region of the image. Compared to non-dense fog regions, dense fog areas have a higher density of dust particles, higher scattered light intensity, larger overall image brightness, and smaller channel difference values. Based on these characteristics, dust and fog image regions are segmented. Image channel difference is defined as: (1) in, This represents the channel map of the input image. For image channel differences, Represents color channels. This represents the maximum color channel value for each region of the image. This represents the minimum color channel value for each region of the image. The gamma transform is used to improve the brightness difference of an image. The brightness of the transformed image is expressed as follows: (2) in, The brightness map after Gamma transformation is shown below. Normalization is performed, as shown in formula (3): (3) in, This is the normalized brightness map. This represents the minimum value in the brightness graph. The value represented by the maximum value in the brightness map indicates that areas with dense fog have higher overall image brightness, and there are channel differences. ; Based on the above priors, the probability distribution function of the dust and fog region in the image is constructed as shown in formula (4): (4) The maximum entropy threshold segmentation method is used to determine the threshold, and the probability distribution function of the entire coal mining face dust and fog image is segmented using this threshold to obtain the dense fog area and the non-dense fog area.

2. The method for clarifying dust and fog images in coal mining faces according to claim 1, characterized in that, Step 2 specifically involves: Step 2.

1. Set the initial lighting map; The initial illumination map of the dust and fog image of the coal mining face was estimated using the Max-RGB method. Let the initial illumination map be... for: (5) Step 2.

2. Solve for the refined lighting map; The optimization objective function that simultaneously preserves the overall structure and edge information of the image is constructed as follows: (6) in, Represents the Frobenius norm. Indicates image structure retention terms. Represents the weight matrix. Represents a detailed lighting map; express Norm; To preserve the overall structure of the lighting map, the initial lighting map obtained in step 2.1 is... And the desired refined lighting map Combining to obtain Simultaneously, in order to preserve the edge information of the image in the illumination map, a weight matrix is ​​constructed. The weight matrix is ​​then compared with the refined illumination map. By combining, we can obtain ; weight matrix Based on the relative total variation model, the weight matrix The expression is: (7) in, Indicated by The neighborhood centered on, The standard deviation is Gaussian kernel function, The gradient symbol, Represents the gradient direction. This represents the initial lighting diagram. for Coordinates in; Represented as: (8) The approximate value is expressed as: (9) in, , These represent the horizontal and vertical directions, respectively. To avoid the denominator being zero during calculation, a number close to zero needs to be added to the denominator. , The objective function is rewritten as follows: (10) The scaling factor is used to balance the two penalty factors. To solve formula (10), we obtain the following based on Lagrange's theorem: (11) remember Rewriting formula (11) in discrete form, and finding an approximate solution to the above equation, the calculation formula is as follows: (12) in, Represents the forward finite partial differential, Represents backward finite partial differential equations. The iteration step size, , These are the refined illumination maps for the nth and (n+1)th estimates, respectively; assuming that in the Nth iteration, formula (6) reaches its minimum value or the maximum number of iterations, This refers to the refined lighting map that is desired. .

3. The method for clarifying dust and fog images in coal mining faces according to claim 2, characterized in that, Step 3 specifically involves: Step 3.

1. Estimation of transmittance values ​​in dense fog areas; Transmittance in atmospheric attenuation models With scene depth and atmospheric scattering coefficient Related, expressed as: (13) Among them, the brightness, saturation and gradient features of the dense fog area image of the coal mining face approximately conform to the color decay model, and the scene depth of this area is linearly represented by gradient, brightness and saturation. Based on the linear image depth model of the color attenuation model, the scene depth of the dense fog region is represented as: ， (14) in, , , The coefficients are linear. This represents the estimation error. Indicates an area of ​​dense fog. , , , These represent the image's depth, brightness, saturation, and gradient information, respectively. Estimate the average depth of dust and fog by selecting a fixed-size window range. for: (15) in, For The central window is N, where N is the number of windows. Indicates the average depth of the window. , , These represent the window's average brightness, average saturation, and average gradient, respectively. In dense fog areas underground, dust and fog density is uneven, and the atmospheric scattering coefficient is related to the dust and fog concentration. The atmospheric scattering coefficient is expressed as: (16) Where a and b are coefficients, The brightness image is represented; the scene depth and atmospheric scattering coefficient are calculated according to formulas (14) and (16) respectively, and then substituted into formula (15) to obtain the transmittance of the fog area. ; Step 3.

2. Estimation of transmittance values ​​in non-dense fog areas; For areas outside of dense fog, the transmittance of these areas is obtained using the dark channel prior. for: (17) in, Represented in pixels A window to the center; This represents the ambient light map for each channel.

4. The method for clarifying dust and fog images in coal mining faces according to claim 3, characterized in that, Step 4 specifically involves: Step 4.

1. Ambient light estimation in dense fog areas; The light distribution in the dense fog area is represented as follows: (18) in, This indicates the ambient light level in areas with dense fog. Indicates the transmittance in areas of dense fog; In step 2, the global illumination map is estimated. Then, the ambient light value in the dense fog area... for: (19) Transmittance in dense fog areas The ambient light value of the dense fog area is obtained by solving step 3.

1. Step 4.

2. Ambient light estimation in non-dense fog areas; Ambient light value in non-dense fog areas for: (20).

5. The method for clarifying dust and fog images in coal mining faces according to claim 4, characterized in that, Step 5 specifically involves: When calculating the global ambient light and transmittance matrix, the probability function for dense fog regions is used. The ambient light values ​​and transmittance of different regions are fused using Alpha to obtain the global ambient light value. and transmittance for: (21) For global transmittance Guided filtering is performed to reduce noise generated during the fusion process while preserving the edges; the low-light dust and fog image is recovered using an atmospheric scattering model, as shown in formula (22); (22) in, Represents the restored, sharpened image; This represents the transmittance correction parameter. Indicates global transmittance and The maximum value in.

Images