Haze image sharpening algorithm based on atmospheric scattering model

Through the haze image clarification algorithm based on the atmospheric scattering model, the atmospheric scattering model is corrected and the areas of dense haze and non-concentrated haze are divided, the transmittance and atmospheric light value are calculated, and the image defog removal in haze environment in the existing technology is solved. The algorithm's color distortion and insufficient robustness achieve more accurate and clear haze image processing.

CN119006325APending Publication Date: 2024-11-22XIAN TECH UNIV
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202411077143.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing technology image defog removal algorithm cannot effectively distinguish different haze environments in haze environments, resulting in color distortion and inaccuracy of clear images, and even failure in severe haze environments, and dark channel prior image defog The algorithm distorts the areas with high brightness and is not robust enough.

Method used

The haze image clarification algorithm based on the atmospheric scattering model is used to correct the atmospheric scattering model, and divide the regions according to the haze concentration, calculate the transmittance and atmospheric light values ​​of dense and non-concentrated haze areas, and combine the global light intensity for fusion processing. , generate clear images.

Benefits of technology

It effectively overcomes the edge effect and noise problems during haze image area segmentation, improves the contrast and clarity of the image, makes the atmospheric light value and transmittance more in line with the real attributes, avoids color distortion, and improves the lucidity of haze image processing Strong and clear.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006325A_ABST
    Figure CN119006325A_ABST
Patent Text Reader

Abstract

The invention discloses a haze image sharpening algorithm based on an atmospheric scattering model, and the algorithm comprises the steps: correcting an atmospheric scattering model through employing the physical characteristics of a haze environment, and obtaining a corrected atmospheric scattering model; an original haze image is converted from an RGB color space to an HIS color space, the haze image is divided into a dense haze area and a non-dense haze area, and the limitation that a dark channel-based image defogging algorithm cannot adapt to an image without a sky area is overcome; calculating the global light intensity of the haze image; respectively calculating the transmissivity and the atmospheric light value of the dense haze area and the non-dense haze area; fusing the transmissivity and the atmospheric light value of the dense haze area and the non-dense haze area to obtain a global transmissivity and a global atmospheric light value respectively; and substituting the obtained global transmissivity and the global atmospheric value into the corrected atmospheric scattering model to obtain a haze image after sharpening processing. And the atmospheric light value and the transmissivity in the image sharpening process better conform to the real attribute of the haze image.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a haze image clarity algorithm based on an atmospheric scattering model. Background Art

[0002] In a haze environment, images collected by various computer vision devices will also be degraded, blurred, distorted, and lost in detail, which will make it impossible for the computer vision system to complete subsequent mid- and high-level visual tasks such as image analysis, target detection, target recognition, and target tracking. However, in real scenes, the concentration of haze and lighting conditions vary in different haze environments. Directly inverting the atmospheric scattering model to achieve the sharpening of degraded images in haze environments will cause the sharpened images to have color distortion and inaccurate sharpening results. In some environments with particularly severe haze pollution, the sharpening algorithm will even fail. This is mainly because most existing image dehazing algorithms do not distinguish between different meteorological environments for ease of processing.

[0003] Among the existing dehazing algorithms, the dark channel prior image dehazing algorithm is prone to color distortion for areas with higher brightness in the image, and there are still problems with the algorithm's universality and robustness when applied to actual haze images. In fact, there are significant differences in the physical properties of fog and haze, and haze is a natural phenomenon that occurs when fog and haze are mixed together; if the dehazing algorithm is used to clarify the haze image, the image will be distorted and the effect of clarifying the haze image will not be achieved; therefore, in order to improve the performance of the haze image clearing algorithm, the physical properties of the haze environment itself must be considered; after clarifying the essential physical causes of image degradation and distortion in the haze environment, the clearing of images with different degrees of haze environment degradation can be achieved in a targeted manner. The classic dark channel prior image dehazing algorithm based on the atmospheric scattering model is used to process a haze image in a real environment; for haze images with unclear sky areas, the dark primary color prior image dehazing algorithm is basically ineffective, and the haze image after clearing has color distortion as a whole, severe halo phenomenon, and too many natural attributes are missing. Summary of the invention

[0004] Purpose of the invention: In order to overcome the deficiencies in the prior art, the present invention provides a haze image clarity algorithm based on an atmospheric scattering model, which segments the haze image into regions according to the haze concentration, obtains the probability density function of the dense haze area, and respectively calculates the atmospheric light value and transmittance of the dense haze area and the non-dense haze area to achieve haze image clarity.

[0005] Technical solution: To achieve the above purpose, a haze image clearing algorithm based on an atmospheric scattering model of the present invention comprises the following steps:

[0006] S1. Using the physical characteristics of the haze environment to correct the atmospheric scattering model, to obtain a corrected atmospheric scattering model;

[0007] S2, converting the original haze image from RGB color space to HIS color space, and dividing the haze image into dense haze area and non-dense haze area according to channel brightness difference;

[0008] S3, calculating the global light intensity of the haze image;

[0009] S4, obtaining the transmittance of the dense haze area according to the atmospheric scattering coefficient and scene depth of the dense haze area, and calculating the atmospheric light value of the dense haze area in combination with the global light intensity; calculating the atmospheric light value of the non-dense haze area and the transmittance of the non-dense haze area according to the global light intensity;

[0010] S5, fusing the transmittance of the dense haze area with the transmittance of the non-dense haze area to obtain a global transmittance, and fusing the atmospheric light value of the dense haze area with the atmospheric light value of the non-dense haze area to obtain a global atmospheric light value;

[0011] S6. Substituting the obtained global transmittance and global atmospheric value into the modified atmospheric scattering model to obtain a cleared haze image.

[0012] Furthermore, the modified atmospheric scattering model is:

[0013] I * (x, y) = J (x, y) t (x, y) + A * (x,y)[1-t(x,y)]

[0014] Where J(x, y) is the cleared target image, and t(x, y) is the transfer function;

[0015] A * (x, y) = Ψ⊙A(x, y)

[0016]

[0017] Wherein, “⊙” represents the symbol of the Hadamard product, λ is the wavelength of light within the wavelength range Ω of visible light, l(λ) is the brightness of the incident light source, c(λ) is the value of the three color channels, c(λ) = {R(λ), G(λ), B(λ)}, N represents the integration interval of the dense haze area; c is any one of the three color channels: red, green, and blue.

[0018] Furthermore, in step S2, the original haze image is converted from the RGB color space to the HIS color space, and the conversion process of the three color channels of red R, green G and blue B in the RGB color space and the three components of hue H, saturation S and brightness I in the HIS color space is as follows:

[0019]

[0020] Based on the characteristics of the haze image, the saturation component S of the HIS color space is adjusted, and the brightness component I of the HIS color space is adjusted; the adjusted saturation S O The calculation process of (x, y) is as follows:

[0021]

[0022] Where S(x, y) is the initial saturation, S min , S max are the minimum and maximum values ​​of the saturation component respectively;

[0023] Get the adjusted brightness I V The calculation process of (x, y) is as follows:

[0024] I V (x, y) = 0.4520·I R (x,y)+0.5121·I G (x,y)+0.0359·I B (x, y)

[0025] In the formula, I R (x, y), I G (x, y), I B (x, y) are the values ​​of the three color components R, G, and B of the haze image respectively.

[0026] Furthermore, in step S2, the channel brightness difference Z of the haze image κ (x, y) is defined as:

[0027]

[0028] The haze image converted to the HIS color space is enhanced by gamma transformation to obtain the gamma-enhanced haze image brightness value δ(x, y), and then normalized to obtain the normalized haze image brightness value. The calculation process is as follows:

[0029]

[0030] Where γ is the gamma parameter, δ min and δmax are the minimum and maximum brightness values ​​of the haze image after gamma enhancement;

[0031] Construct the distribution probability density function of the dense haze area in the haze image

[0032]

[0033] Use the maximum entropy threshold segmentation method to select the segmentation threshold F T ,Based on the distribution probability density function of the dense haze area of ​​the haze image, the haze image is segmented to obtain the dense haze area and the non-dense haze area.

[0034] Furthermore, in step S3, the haze image is divided into four sub-block areas, and the score of each area is calculated. Determine whether the sub-block area with the highest score belongs to the dense haze area;

[0035]

[0036] In the formula, To calculate the average grayscale of each area, n is the number of segmentation, i = 1 represents a dense haze area, and i = 2 represents a non-dense haze area;

[0037] When the area with the highest score belongs to the non-dense haze area, the segmented area is output; when the area with the highest score belongs to the dense haze area, the area with the highest score is weighted by the distribution probability density function of the dense haze area, and the segmented area is output;

[0038] The segmented area is divided into four sub-block areas again, and the scores of the re-segmented areas are calculated, and it is determined whether the area with the highest score belongs to the dense haze area; the cycle is repeated until the segmentation termination condition is reached, the segmentation is terminated, and the final segmented area X is output. over ; The last segmentation area X over The first 10% brightness pixels are taken as the global light intensity M(x, y); the segmentation termination condition is:

[0039]

[0040] In the formula, is the maximum grayscale mean of the area with the highest segmentation score for the nth time, It is the second largest grayscale mean of the area with the highest segmentation score for the nth time.

[0041] Furthermore, when the two final segmented regions are obtained, the scores of the last two regions are calculated based on the average grayscale value and the average gradient value. And compare; the last segmentation area with the highest score is taken as the last segmentation area X over ;

[0042]

[0043] In the formula, Represents the average grayscale value of the region. Represents the average gradient of the region.

[0044] Furthermore, in step S4, based on the principle of the sunlight physical model, the atmospheric scattering coefficient of the dense haze area is obtained, and the grayscale information of the illumination map is integrated to obtain the atmospheric scattering coefficient β1(x, y) of the dense haze area:

[0045]

[0046] In the formula, and τ are fixed coefficients, g(x, y) is the grayscale value of the illumination map corresponding to the dense haze area;

[0047] A linear model is established and the gradient information of the dense haze image area is integrated to calculate the scene depth of the dense haze area:

[0048] d1(x,y)=α1I(x,y)+α2S(x,y)+α3G(x,y)+ε

[0049] Where I(x, y), S(x, y), G(x, y) represent the brightness, saturation and gradient information of the image in the dense haze area, α1, α2, α3 are linear coefficients, and ε is the system random error;

[0050] Substitute the atmospheric scattering coefficient β1(x, y) of the dense haze area and the scene depth d1(x, y) of the dense haze area into the transmittance calculation formula to obtain the transmittance t1(x, y) of the dense haze area; combine the transmittance of the dense haze area with the global light intensity to calculate the atmospheric light value A1(x, y) of the dense haze area.

[0051] Furthermore, in step S4, the calculation process of the atmospheric light value in the non-dense haze area is as follows;

[0052] A2(x, y) = M(x, y)

[0053] The transmittance t2(x, y) of the non-dense haze area is obtained by the dark channel prior algorithm:

[0054]

[0055] Where C is any channel of the RGB color channel, A2(x, y) is the atmospheric light value in the non-dense haze area, ω(x, y) represents a window centered at (x, y) in the non-dense haze area, Refers to areas without heavy haze.

[0056] Beneficial effects: The haze image clearing algorithm based on the atmospheric scattering model of the present invention converts the haze image from the RGB color space to the HIS color space, thereby enhancing the contrast of the haze image; the haze image is divided into a dense haze area and a non-dense haze area, thereby overcoming the limitation that the sky part of the image must be obtained when obtaining the atmospheric light value based on the dark channel image dehazing algorithm, and the image cannot be adapted to the image without the sky area; firstly, the global light intensity is calculated based on the divided dense haze area and the non-dense haze area, and the transmittance and atmospheric light value of the dense haze area and the non-dense haze area are calculated based on the global light intensity, thereby making the atmospheric light value and transmittance in the image clearing process more consistent with the real attributes of the haze image; the transmittance of the dense haze area and the non-dense haze area are fused, and the atmospheric light values ​​of the dense haze area and the non-dense haze area are fused, thereby effectively avoiding the edge effect when the haze image area is segmented, and at the same time suppressing the noise generated by the global transmittance matrix. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is the flow chart of the haze image clarity algorithm based on the atmospheric scattering model;

[0058] Figure 2 Different color spaces of haze images and corresponding color histograms;

[0059] Figure 3 This is the haze distribution map of different concentrations of haze images;

[0060] Figure 4 It is the result map of haze image area segmentation and the probability density distribution map of dense haze area;

[0061] Figure 5 A comparison diagram of the transmittance matrix effect between the method of the present invention and the dark pass prior algorithm;

[0062] Figure 6 This is a comparison chart of the clarity effects of different algorithms for artificially synthesized haze images;

[0063] Figure 7 This is a comparison chart of the processing results of different algorithms for haze image clarity in a light haze environment;

[0064] Figure 8 This is a comparison chart of the processing results of different algorithms for haze image clarity in a moderate haze environment;

[0065] Fig. 9 This is a comparison chart of the haze image clearing processing results of different algorithms in a yellow haze scene under real environment;

[0066] Fig.10 This is a comparison chart of the processing results of different algorithms for haze image clarity in a real environment with severe haze;

[0067] Fig.11 for Fig. 9 Histogram of processing time of different algorithms for images under moderate to severe haze pollution. DETAILED DESCRIPTION

[0068] The present invention will be further described below in conjunction with the accompanying drawings.

[0069] like Figure 1 As shown, a haze image clearing algorithm based on an atmospheric scattering model includes the following steps:

[0070] S1. Using the physical characteristics of the haze environment to correct the atmospheric scattering model, to obtain a corrected atmospheric scattering model;

[0071] S2, converting the original haze image from RGB color space to HIS color space, and dividing the haze image into dense haze area and non-dense haze area according to channel brightness difference based on the maximum entropy threshold segmentation method;

[0072] S3, using quadtree segmentation algorithm to calculate the global light intensity of the haze image;

[0073] S4, obtaining the transmittance of the dense haze area according to the atmospheric scattering coefficient and scene depth of the dense haze area, and calculating the atmospheric light value of the dense haze area in combination with the global light intensity; calculating the atmospheric light value of the non-dense haze area and the transmittance of the non-dense haze area according to the global light intensity;

[0074] S5, using Alpha fusion to fuse the transmittance of the dense haze area with the transmittance of the non-dense haze area to obtain the global transmittance, and fuse the atmospheric light value of the dense haze area with the atmospheric light value of the non-dense haze area to obtain the global atmospheric light value;

[0075] S6. Substituting the obtained global transmittance and global atmospheric value into the modified atmospheric scattering model to obtain a cleared haze image.

[0076] In the field of image defogging, the most commonly used physical model for defogging algorithms based on atmospheric scattering models is the traditional atmospheric scattering model:

[0077] I(x,y)=J(x,y)t(x,y)+A[1-t(x,y)]

[0078] In the formula, I(x, y) is the image acquired by the detection system, that is, the input image to be cleared in the haze environment; A is the atmospheric light value at infinity; J(x, y) is the target radiation light, that is, the cleared target image; t(x, y) is the transfer function, that is, the transmittance. The transmittance is related to the atmospheric scattering coefficient β and the scene depth d(x, y), and the expression is:

[0079] t(x, y) = e-βd(x,y)

[0080] The expression of the clear haze image J(x, y) is obtained as follows:

[0081]

[0082] Haze image sharpening algorithms based on atmospheric scattering models are usually based on the ideal assumptions that the size of atmospheric particles is uniform and the concentration distribution of atmospheric media is uniform, but the actual haze weather is quite different from the above ideal assumptions. Since the physical causes of haze images are not considered, the image dehazing algorithms based on the atmospheric scattering model do not achieve ideal sharpening effects when processing haze images in real environments, and cannot adaptively process low-quality images affected by haze to varying degrees. In order to solve the problem that the existing image dehazing algorithms based on the atmospheric scattering model do not achieve ideal results when processing haze images in real environments, starting from the physical characteristics of the haze environment, the imaging process in the haze environment is analyzed to find out the essential causes of image distortion in the haze environment; the atmospheric scattering model is corrected based on the physical characteristics of the haze environment, thereby improving the performance of the haze image sharpening algorithm based on the corrected atmospheric scattering model.

[0083] Since the atmospheric scattering model is based on the ideal assumption that the size of atmospheric particles is uniform, the composition of atmospheric particles under different weather environments is first compared as shown in Table 1. It can be seen that there are significant differences in the composition of atmospheric particles in different weather types. For image sharpening processing in haze weather environments, the physical properties of the two environments of "fog" and "haze" should be considered. Scattering in the atmosphere includes Rayleigh scattering, Mie scattering and non-selective scattering. According to the composition of atmospheric particles in "fog" and "haze", their composition can be unified as atmospheric aerosols with a radius of 10-2 to 10μm. The wavelength range of visible light is 0.38 to 0.78μm, and the types of scattering in haze environments include Rayleigh scattering and Mie scattering.

[0084] Table 1 Comparison of atmospheric particle composition in different meteorological types

[0085]

[0086]

[0087] The specific components of atmospheric particles in haze weather mainly include coal particles, heavy metal particles, sulfates, nitrates and other air pollutants, which lead to a sharp increase in the concentration of Eigen nucleus mode aerosols with a particle radius less than 0.1 and accumulation mode aerosols with a particle radius between 0.1 and 1.0; and the wavelengths of light corresponding to the scattering of different particle components are different. Mie scattering in haze environment causes the brightness of haze images in the distance to decrease. The mutual influence of Rayleigh scattering between different types of atmospheric particles causes the blue and purple light with shorter wavelengths to be unable to enter the imaging device, and the red and yellow light with longer wavelengths are retained more, resulting in the color distortion and dark color of the sky area in the haze image processed by the image defogging algorithm based on the atmospheric scattering model. According to the physical characteristics of the haze environment, the wavelengths of light corresponding to the scattering of atmospheric particles of different components are different. In the atmospheric scattering model, the atmospheric light comes from the direct and diffuse reflection of sunlight. The difference between the two causes the haze image processed by the image defogging algorithm based on the atmospheric scattering model to still have color distortion; therefore, the atmospheric scattering model in haze environment is corrected based on atmospheric light parameters.

[0088] According to the surface scattering model, the image brightness is defined as the integral of the light source brightness, surface reflectance, and color channels;

[0089]

[0090] In the formula, λ is the wavelength of light within the wavelength range Ω of visible light, l(λ) is the brightness of the incident light source, c(λ) is the value of the three color channels, c(λ) = {R(λ), G(λ), B(λ)}, s(λ, x) is the surface reflection of the object at position x in the image. The expression of the light source illumination e in the transmittance is as follows:

[0091]

[0092] For an RGB image, c(λ) is the value of the three color channels, simplified to the vector [R(λ), G(λ), B(λ)], defining ζ r , g and b are the three colors of atmospheric light, and the calculation process is as follows:

[0093]

[0094] Where c is any one of the three color channels: red, green, and blue, and N represents the integral interval of the dense haze area.

[0095] Because the human eye is more sensitive to monochromatic green light, in order to estimate the degree of color distortion, g The haze image color correction vector Ψ is obtained as the standard:

[0096]

[0097] By calculating the Haadman product of the color distortion correction vector Ψ and the original atmospheric light value A, we can get the atmospheric light value A without color shift. * ; In the original atmospheric scattering model, the atmospheric light value A often takes a global constant value, but for the real haze environment, affected by the composition and concentration of atmospheric particles, the images taken in the haze environment have uneven lighting. Therefore, in the modified atmospheric scattering model, the constant atmospheric light value is expanded to the atmospheric light matrix A(x, y); when calculating the color distortion correction vector, the difference between each channel is calculated. If the haze concentration is low and the image color distortion is not serious, it will not cause over-correction, which effectively ensures the robustness of the modified atmospheric scattering model for images in haze environments of different concentrations; therefore, the atmospheric light matrix A without color shift is obtained * (x, y) is:

[0098] A * (x, y) = Ψ⊙A(x, y)

[0099] Wherein, “⊙” represents the symbol of the Hartmann product.

[0100] The modified atmospheric scattering model is:

[0101] I * (x, y) = J (x, y) t (x, y) + A * (x, y)[1-t(x, y)].

[0102] At present, most image dehazing methods, including dark channel prior methods, are performed in RGB color space. Due to the overlap of spectra, the three color channels of RGB space will be coupled, and the dehazed images of some dense haze images will have the problem of oversaturation. HSI color space describes color through three components: hue, saturation and brightness, which can make the color description more natural and intuitive, and easy to understand and describe color from the perspective of human visual characteristics. In order to avoid the problem of coupling of the three color channels R, G, and B in RGB space due to the overlap of spectra in most existing image dehazing algorithms, the initial haze image is transferred to the more intuitive color model HSI color space that is closer to the visual characteristics of the human eye, and the shortcomings of HSI color space are adjusted.

[0103] In step S2, in order to make the haze image more consistent with the visual effect of human eyes, the original haze image is converted from the RGB color space to the HIS color space. The conversion process of the three color channels of red R, green G and blue B in the RGB color space and the three components of hue H, saturation S and brightness I in the HIS color space is as follows:

[0104]

[0105] Although the maximum value of each channel of saturation S is suppressed, the problem of over-saturation still occurs. Therefore, an adaptive saturation enhancement method is proposed for the saturation component of the HSI color space. Based on the characteristics of the haze image, the saturation component S of the HSI color space is adjusted to achieve global saturation adaptive enhancement of the haze image. The adjusted saturation S O The calculation process of (x, y) is as follows:

[0106]

[0107] Where S(x, y) is the initial saturation, S min , S max are the minimum and maximum values ​​of the saturation component respectively;

[0108] The brightness component I of the HSI color space is the color component mean of the three RGB color channels. However, due to the Rayleigh scattering of atmospheric particles, the contribution rates of the three color channels in the image under haze environment are not the same during the imaging process. The method of taking the color component mean cannot reflect the true brightness attribute of the haze image. For the brightness component I of the HSI color space, according to color theory, the human eye has different subjective feelings about the brightness of monochromatic light of the same intensity. Taking the three primary colors of RGB as an example, the human eye feels that green light has the strongest brightness, red light is about 1 / 2 of green light, and blue light is the weakest, about 1 / 3 of red light; assuming that the visual light intensity of green light is 1, the intensity of red light can only be 1 / 2 of the original intensity, and the intensity of blue light can only be 1 / 6 of the original intensity. Therefore, the coefficients of the three primary colors of RGB in the brightness component are determined by combining the difference in the human eye's perception of monochromatic light of the same intensity and the relationship between the scattering rate of each color light and the wavelength when atmospheric particles scatter under haze environment. The brightness component I adjusts the weight coefficients of the contribution of the three RGB channels to the brightness according to the characteristics of the haze image; the brightness component I of the HIS color space is adjusted to obtain the adjusted brightness I V The calculation process of (x, y) is as follows:

[0109] I V (x, y) = 0.4520·I R (x,y)+0.5121·I G (x,y)+0.0359·I B (x, y)

[0110] In the formula, I R (x, y), I G (x, y), I B (x, y) are the values ​​of the three color components R, G, and B of the haze image respectively.

[0111] like Figure 2 As shown in the figure, the visual effects of a haze image taken in a real environment in the RGB color space of the image itself and the improved HSI color space, as well as the histogram distribution of the corresponding three color channels; Figure 2 (a) is the visual effect of the RGB color space of the haze image itself. Figure 2 (b) is the visual effect diagram of the improved HSI color space. Figure 2 (c) is the original haze image and the three-channel histogram distribution in RGB color space. Figure 2 (d) is the image converted to the improved HSI color space in this paper and the corresponding three-channel histogram distribution. The original image is darker in color due to haze interference, and the color histogram is concentrated in the grayscale range of 150-200; in the improved HSI color space, compared with the original image, the histograms of the three color channels of the image are relatively evenly distributed in the entire grayscale range, and the brightness of the haze image is significantly improved. The blur and halo problems in the image need further processing to obtain a clearer image.

[0112] In step S2, for the images collected in some haze environments, the haze concentration distribution has obvious unevenness. Generally, the haze distribution density in the distant view of the image is relatively large, which is defined as a dense haze area; the haze distribution density in the near view of the image is relatively small, which is defined as a non-dense haze area. Figure 3 As shown, Figure 3 (a) and Figure 3 (b) In the two images, A is a dense haze area, and B is a non-dense haze area. Since the haze image has been converted to the improved HSI color space, the areas with higher haze concentration and those with lower haze concentration are more obvious in visual effect. Therefore, the segmentation of dense haze area and non-dense haze area is achieved based on the channel brightness difference of the haze image; in the area with higher haze concentration, the distribution density of atmospheric particles is larger, the scattering effect is stronger, the image brightness channel difference value is smaller, and the brightness value is larger. The channel brightness difference Z of the haze image is κ (x, y) is defined as:

[0113]

[0114] Since the channel brightness difference in the dense haze area is very small, it is approximated to 0, so Z κ (x, y)→0; In order to further improve the accuracy of haze image region segmentation; the haze image converted to the HIS color space uses gamma transformation to enhance the channel brightness difference value, and obtains the gamma enhanced haze image brightness value δ(x, y). The calculation process is as follows:

[0115]

[0116] In the formula, γ is the gamma parameter, and combined with the test results, we get γ = 0.8.

[0117] In order to unify the statistical distribution of the brightness of the haze image, the gamma-enhanced image brightness value δ(x, y) is normalized to obtain the normalized haze image brightness value

[0118]

[0119] In the formula, δ min and δ max They are the minimum and maximum brightness values ​​of the haze image after gamma enhancement.

[0120] The overall brightness value of the dense haze image area is higher, and the distribution probability density function of the dense haze area of ​​the haze image is constructed.

[0121]

[0122] Use the maximum entropy threshold segmentation method to select the segmentation threshold F T , based on the distribution probability density function of the dense haze area of ​​the haze image, the haze image is segmented to obtain the dense haze area and the non-dense haze area. Figure 4 As shown, the segmentation result of the dense haze area obtained by using the segmentation method of the present invention for an image with uneven haze concentration distribution and the corresponding three-dimensional probability distribution of the dense haze area; Figure 4 (a) is a haze image. Figure 4 (b) The result of haze image region segmentation. Figure 4 (c) Probability density distribution of dense haze area. From the segmentation results, we can see that the dense haze area segmented by the method proposed in the present invention basically conforms to the division of dense haze area by human eyes. Figure 4 (c) It can be seen that the grayscale level of the dense haze image area is higher, mostly concentrated in the yellow area. Since the essence of regional segmentation of haze images is based on the segmentation of bright and dark areas of haze images, there will be a few non-dense haze areas with high brightness in the image defined as dense haze areas, but this will not affect the subsequent image defogging results.

[0123] In step S3, the traditional dark channel prior image dehazing algorithm usually estimates the global light intensity by counting the pixels with the top 0.1% brightness from the image, but this method will affect the accuracy of the atmospheric light value when there are white or high-brightness objects in the image. Since the haze image has been divided into dense haze areas and non-dense haze areas according to the channel brightness difference based on the maximum entropy threshold segmentation method, the divided dense haze areas and non-dense haze areas use the quadtree segmentation method to obtain the global light intensity. The dense haze area of ​​the haze image is defined as The non-dense haze area is defined as Divide the haze image into four sub-block areas and calculate the score of each area Determine whether the sub-block area with the highest score belongs to the dense haze area;

[0124]

[0125] In the formula, To calculate the average grayscale of each area, n is the number of segmentation, i = 1 represents a dense haze area, and i = 2 represents a non-dense haze area;

[0126] When the area with the highest score belongs to the non-dense haze area, there is Output segmented area; when the area with the highest score belongs to the dense haze area, there is The area with the highest score is weighted by the distribution probability density function of the dense haze area, and Compare the sizes and output the segmented areas;

[0127] The segmented area is divided into four sub-block areas again, and the scores of the re-segmented areas are calculated, and it is determined whether the area with the highest score belongs to the dense haze area; the cycle is repeated until the segmentation termination condition is reached, the segmentation is terminated, and the final segmented area X is output. over The essence of global light intensity is to find the pixel point in the current block that is closest to the highest light intensity (255, 255, 255). The test experiment found that the first 10% of the brightness pixels in the final segmentation area are closest to (255, 255, 255). Therefore, the final segmentation area X over The first 10% brightness pixels of the segmented area are taken as the global light intensity M(x, y); for the nth segmented area, the grayscale average value of the segmented area is recorded as The segmentation termination condition is that the difference between the maximum grayscale mean and the second largest grayscale mean of the segmented area is less than or equal to the segmentation threshold F between the dense haze area and the non-dense haze area. T , the expression is:

[0128]

[0129] In the formula, is the maximum grayscale mean of the area with the highest segmentation score for the nth time, It is the second largest grayscale mean of the area with the highest segmentation score for the nth time.

[0130] When the two final segmented regions are obtained, the scores of the last two regions are calculated based on the average gray value and the average gradient value. And compare; the last segmentation area with the highest score is taken as the last segmentation area X over ;

[0131]

[0132] In the formula, Represents the average grayscale value of the region. Represents the average gradient of the region.

[0133] In the step S4, in the atmospheric scattering model, the transmittance is related to the atmospheric scattering coefficient and the scene depth. Since the atmospheric scattering model is based on the ideal assumption that all media are uniformly distributed, β is often taken as a fixed constant value for convenience during calculation. In the actual haze environment, the atmospheric particles are not completely uniformly distributed. The atmospheric scattering coefficient is related to the distribution density of the atmospheric particles in the haze. The greater the density, the greater the brightness of the image. Based on the principle of the daylight physics model, the atmospheric scattering coefficient of the dense haze area is obtained, and the grayscale information of the illumination map is integrated to obtain the atmospheric scattering coefficient β1(x, y) of the dense haze area:

[0134]

[0135] In the formula, and τ are fixed coefficients, τ=0.25; g(x, y) is the grayscale value of the illumination map corresponding to the dense haze area;

[0136] The scene depth characteristics of the dense haze area conform to the color attenuation model. A linear model is established and the gradient information of the dense haze image area is integrated to calculate the scene depth of the dense haze area:

[0137] d1(x,y)=α1I(x,y)+α2S(x,y)+α3G(x,y)+ε

[0138] In the formula, I(x, y), S(x, y), G(x, y) represent the brightness, saturation and gradient information of the image in the dense haze area, α1, α2, α3 are linear coefficients, ε is the system random error, ε = 0.415. The coefficient values ​​before the three parameters of brightness, saturation and gradient are obtained by training data through supervised learning method; when determining the corresponding coefficients, 500 haze-free images in the RESIDE dataset are used to generate training samples; when training data, the joint conditional distribution is used to maximize the probability scene depth to obtain the optimal parameter component coefficients; the empirical values ​​of the three coefficients are α1 = 1.036, α2 = -0.578, and α3 = 0.132.

[0139] Substitute the atmospheric scattering coefficient β1(x, y) and the scene depth d1(x, y) of the dense haze area into the calculation formula of the transmittance t(x, y) to obtain the transmittance t1(x, y) of the dense haze area; the light intensity in the dense haze area is relatively high, and the relative density of atmospheric particles of various components is relatively large. The atmospheric light value that decays with the scene depth is the main source of illumination; the attenuation process of the atmospheric light value can be expressed by the transmittance in the atmospheric scattering model, then the light intensity in the dense haze area can be expressed as M(x, y) = A1(x, y)t1(x, y), and the calculation formula of the atmospheric light value in the dense haze area is derived; the transmittance and global light intensity of the dense haze area are combined to calculate the atmospheric light value A1(x, y) in the dense haze area;

[0140]

[0141] In step S4, for the non-dense haze area, the ambient light intensity is small, and the transmittance t2(x, y) of the non-dense haze area is obtained by the dark channel priori algorithm:

[0142]

[0143] Where C is any channel of the RGB color channel, A2(x, y) is the atmospheric light value in the non-dense haze area, ω(x, y) represents a window centered at (x, y) in the non-dense haze area, Refers to areas without heavy haze.

[0144] The scattering effect of atmospheric particles determines the ambient light in non-dense haze areas. Therefore, the atmospheric light value in non-dense haze areas can be approximated as the global luminosity. The calculation process of the atmospheric light value in non-dense haze areas is as follows;

[0145] A2(x, y) = M(x, y)

[0146] Substituting the obtained atmospheric light value of the non-dense haze area into the transmittance t2(x, y) calculation formula of the non-dense haze area in reverse, the transmittance t2(x, y) of the non-dense haze area can be obtained.

[0147] In step S5, the transmittance t1(x, y) of the dense haze area and the transmittance t2(x, y) of the non-dense haze area are fused by Alpha fusion to obtain the global transmittance t(x, y) of the haze image, and the atmospheric light value A1(x, y) of the dense haze area and the atmospheric light value A2(x, y) of the non-dense haze area are fused to obtain the global atmospheric light A(x, y) of the haze image; the fusion calculation process is as follows:

[0148]

[0149] After obtaining the global atmospheric light A(x, y), the haze image color correction vector Ψ is used to correct it, and the final corrected global atmospheric light matrix A is obtained. * (x, y); In order to reduce the influence of noise introduced when calculating transmittance, the global transmittance matrix is ​​filtered using the fast non-local mean filtering algorithm, which not only effectively removes the image noise introduced in the calculation process, but also effectively maintains the edge and detail information of the haze image transmittance matrix. The global transmittance after fast non-local mean filtering is still recorded as t(x, y). Figure 5 As shown, Figure 4 The final transmittance matrix effects of the haze image obtained by the method of the present invention and the dark channel priori algorithm are compared; Figure 5 (a) is the global transmittance matrix effect diagram obtained by the dark channel prior algorithm. Figure 5 (b) is the effect diagram of the global transmittance matrix obtained by the algorithm of the present invention. Compared with the dark channel prior algorithm, the global transmittance matrix obtained by the method of the present invention is smoother overall. * Substituting the global transmittance t(x, y) and the fast non-local mean filtering processed global transmittance t(x, y) into the modified atmospheric scattering model, the haze image after clarity processing can be solved.

[0150] Example

[0151] In order to verify the performance of the atmospheric scattering model corrected for the haze environment and the haze image clearing algorithm proposed based on the corrected atmospheric scattering model, the present invention carried out a haze image clearing experiment using artificially synthesized haze image data with reference in the public image dehazing dataset RESIDE, REalisticSingle Image Dehazing and various types of real haze image data without reference. In order to comprehensively and objectively evaluate the performance of the proposed haze image clearing algorithm, the haze image clearing effects of the algorithm proposed in the present invention and several classic algorithms in the field of image dehazing and the latest developed algorithms in recent years are compared; these algorithms are respectively the literature Visibility in bad weather from a single image, in which the method proposed is defined as the first algorithm; the literature Single image haze removal using dark channel prior, in which the method proposed is defined as the second algorithm; the literature Efficient Image Dehazing with Boundary Constraint and Contextual Regularization, in which the method proposed is defined as the third algorithm; the literature Haze removal method for natural restoration of images with sky, in which the method proposed is defined as the fourth algorithm; the literature Self-Augnented Unpaired Image Dehazing via Density and Depth Decomposition, in which the method proposed is defined as the fifth algorithm. Since the main purpose of this study is to improve the robustness and clarity performance of the image dehazing algorithm based on the atmospheric scattering model for image processing in different degrees of haze environments, a comparative experiment on real haze image data was conducted, and clarity experiments on light haze, moderate haze, yellow haze scenes and heavy haze were carried out respectively, and the experimental results were compared and analyzed using objective evaluation numerical indicators.

[0152] like Figure 6As shown in the figure, the comparison of the clearing effect of the algorithm in this paper and other comparative algorithms on the artificially synthesized haze images with reference in the RESIDE data set; Figure a is a clear haze-free image, Figure a is an artificially synthesized haze image, Figure c is the image after being cleared by the first algorithm, Figure d is the image after being cleared by the second algorithm, Figure e is the image after being cleared by the third algorithm, Figure f is the image after being cleared by the fourth algorithm, Figure g is the image after being cleared by the fifth algorithm, and Figure h is the image after being cleared by the algorithm of the present invention. The first algorithm has a darker overall color and more serious overall color distortion for the clearing results of different haze images. The dark channel prior algorithm in the second algorithm has a better defogging effect for haze images with non-white sky areas, but the color of the non-sky area in the image is darker; for images with a large amount of white sky area in the image, there is a more obvious distortion. The haze image after being cleared by the third algorithm is better than the first algorithm in visual effect, but there are sporadic mosaic phenomena in the local part of the image. The haze image after being cleared by the fourth algorithm also has the problem of dark brightness in the non-sky area and insufficient image saturation. Compared with the original clear image without haze, the haze image clarified by the fifth algorithm has better restored the original information of the image, but the darker objects in the image, such as "trees" and "vegetation", still have the problem of insufficient brightness. In contrast, the algorithm of the present invention not only effectively realizes image defogging for the four artificially synthesized haze images in the experiment, but also the brightness and color saturation of the image are more in line with the visual characteristics of the human eye; at the same time, compared with the original clear image without haze pollution, it also realizes image enhancement on the basis of retaining the original information of the image.

[0153] In order to more objectively evaluate the clearing effect of different algorithms on the synthetic haze images, the reference evaluation indicators Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM), and the non-reference evaluation indicators Standard Deviation (SD), Average Gradient (AG), and Information Entropy (En) were used to statistically analyze the images processed by the comparison algorithms. The statistical results are shown in Table 2. The reference evaluation indicators Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) can reflect the distortion of the haze image after the clearing process compared with the clear image without haze pollution. The larger the values ​​of the two indicators, the better the performance of the corresponding haze image clearing algorithm. In the comparison experiment, the performance of the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) of the comparison algorithms is basically consistent with the visual subjective evaluation experiment results. The Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) of the proposed algorithm are the largest for the four test images. The Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM) of the image processed by the method in reference

[11] are the worst. The standard deviation (SD) of the non-reference numerical evaluation index can reflect the grayscale distribution and contrast of the clear haze image; the average gradient (AG) reflects the details and texture information of the clear haze image; the information entropy (En) can measure the amount of information contained in the haze image after the clearing process. The larger the value of the above three non-reference numerical evaluation indicators, the better the algorithm performance. Among the tested images, the standard deviation of the image processed by the algorithm of the present invention is the largest, which is twice that of the first algorithm. In terms of the two indicators of average gradient and information entropy, the method of the present invention also has the best performance among several comparison algorithms.

[0154] Table 2 Comparison of objective evaluation indicators of different algorithms for the clarity of artificially synthesized haze images

[0155]

[0156] like Figure 7-10 As shown in Figure 2, in order to further verify the performance of the proposed algorithm in this paper for the clarity of real environment haze images, a comparative experiment was conducted using haze images in real scenes; Figure 7-10 As shown, Figure a is a haze image in a real scene, Figure b is an image after being sharpened by the first algorithm, Figure c is an image after being sharpened by the second algorithm, Figure d is an image after being sharpened by the third algorithm, Figure e is an image after being sharpened by the fourth algorithm, Figure f is an image after being sharpened by the fifth algorithm, and Figure g is an image after being sharpened by the algorithm of the present invention. Figure 7 As shown in the figure, the comparison of the processing results of different algorithms for haze image clarity in a real environment with mild haze is shown in the figure. Figure 8 As shown in the figure, the processing results of different algorithms for haze image clarity in a moderate haze environment are compared. Fig. 9 As shown in the figure, the processing results of haze image clarity of different algorithms in real environment are compared. Fig.10As shown in the figure, the processing results of different algorithms for haze image clarity in a real environment with heavy haze are compared.

[0157] like Figure 7 As shown in Figure 2, for the light haze image in the real scene, the processing effects of the first algorithm and the second algorithm are the least ideal. The light haze image after being clarified by the first algorithm has a darker overall color and loses too much natural attributes. The light haze pollution image processed by the second algorithm has overexposure in the sky area and serious color cast problems, such as Figure 7 As shown, Figure 7 (c1) and Figure 7 (c2). The light haze image processed by the third algorithm has obvious halo problem. The light haze image cleared by the fourth algorithm also has color cast problem, such as Figure 7 shown Figure 7 (e2). In terms of the clarity of images with light haze, the fifth algorithm and the algorithm of the present invention are relatively ideal for processing images with light haze. The fifth algorithm is a deep learning method that relies on a large amount of data. The processing results of the method in this paper are more in line with the visual characteristics of the human eye and have a good clarity effect on different images with light haze. Figure 8 As shown in the figure, for the moderate haze scene, we can see that the visual performance of several algorithms is similar to that of the slight haze scene.

[0158] like Fig. 9 As shown in Figure 2, for heavy haze scenes, the methods of the first and second algorithms are basically ineffective, and the results of haze image clearing are extremely unsatisfactory, such as Fig. 9 shown Fig. 9 (b3) and Fig. 9 (c1), the processed image has a more serious degradation problem than the original haze image; the heavy haze image processed by the third algorithm also has serious halo and noise problems; the overall saturation of the heavy haze image processed by the fourth algorithm is too low; the fifth algorithm has achieved a good clarity of the heavy haze image, but the image still has the problem of too low brightness. Compared with the original heavy haze image and the clarity effect of several other algorithms, the algorithm of the present invention not only better removes the interference of heavy haze in the image, but also improves the overall brightness and saturation of the image, and the image visual effect is more natural.

[0159] like Fig.10 As shown in the figure, for the yellow haze scene in the haze, the processing effects of the first algorithm, the second algorithm, the third algorithm and the fourth algorithm are not particularly ideal, and there are problems such as color cast and uneven brightness. Although the fifth algorithm effectively eliminates the color cast problem for the yellow haze image, some images have halos, such as Fig.10 shown Fig.10(f2). In contrast, the algorithm of the present invention has strong universality and robustness for slight haze, moderate haze, severe haze and yellow haze images in real scenes.

[0160] For the objective evaluation of the clearing effect of haze images in different types of real environments, the present invention uses single numerical indicators without reference, such as information entropy, average gradient, and standard deviation, to evaluate the performance of different algorithms. The specific numerical indicator results are shown in Table 3. By analyzing and comparing the overall numerical indicators, the algorithm of the present invention has higher results for various types of haze images, especially for heavy haze and yellow haze images, the standard deviation of the algorithm of the present invention is 2 to 3 times that of the first algorithm, and 1.3 to 1.5 times that of the dark primary color prior method in the second algorithm; this shows that the algorithm of the present invention has relatively excellent clearing performance for haze pollution images of different degrees and types.

[0161] Table 3 Comparison of objective evaluation indicators of the clarity effect of different algorithms for different types of haze images in real scenes

[0162]

[0163] like Fig.11 As shown, the present invention compares the algorithm of the present invention with the first algorithm, the second algorithm, the third algorithm and the fourth algorithm. Fig. 9 The processing time of three images under moderate to severe haze pollution is shown in the following statistics: Fig.11 As shown; Fig.11 The pixel sizes of the three real-life haze images are 640×480, 1440×1080 and 400×300. Among the comparison algorithms, the fourth algorithm has the fastest processing speed, and the processing speed of the algorithm of the present invention is second only to the fourth algorithm, with an average processing time difference of only 0.11s from the fourth algorithm. Since the main purpose of the present invention is to improve the universality and robustness of the haze image clearing algorithm, the algorithm of the present invention has certain advancement and superiority in processing images with different degrees and types of haze pollution by the present invention.

[0164] The above is only a description of the preferred embodiments of the present invention. Ordinary technicians in this technical field can make several modifications and optimizations based on the above disclosure without departing from the above basic principles. These improvements and optimizations should be regarded as the understood protection scope of the present invention.

Claims

1. A haze image clarity algorithm based on atmospheric scattering model, Features: The following steps are involved: S1. Using the physical characteristics of the haze environment to correct the atmospheric scattering model, to obtain a corrected atmospheric scattering model; S2, converting the original haze image from RGB color space to HIS color space, and dividing the haze image into dense haze area and non-dense haze area according to channel brightness difference; S3, calculating the global light intensity of the haze image; S4, obtaining the transmittance of the dense haze area according to the atmospheric scattering coefficient and scene depth of the dense haze area, and calculating the atmospheric light value of the dense haze area in combination with the global light intensity; The atmospheric light value and transmittance of non-dense haze areas are calculated based on the global light intensity; S5, fusing the transmittance of the dense haze area with the transmittance of the non-dense haze area to obtain a global transmittance, and fusing the atmospheric light value of the dense haze area with the atmospheric light value of the non-dense haze area to obtain a global atmospheric light value; S6. Substituting the obtained global transmittance and global atmospheric value into the modified atmospheric scattering model to obtain a cleared haze image.

2. According to the haze image clearing algorithm based on the atmospheric scattering model described in claim 1, Features: The modified atmospheric scattering model is: I*(x,y)=J(x,y)t(x,y)+A * (x,y)[1-t(x,y)] Where J(x, y) is the cleared target image, and t(x, y) is the transfer function; TO * (x, y)=Ψ⊙A(x, y) Where "⊙" represents the symbol of the Hadamard product, λ is the wavelength of light within the wavelength range Ω of visible light, l(λ) is the brightness of the incident light source, c(λ) is the value of the three color channels, c(λ) = {R(λ), G(λ), B(λ)}, N represents the integration interval of the dense haze area; c is any one of the three color channels: red, green, and blue.

3. According to the haze image clearing algorithm based on the atmospheric scattering model described in claim 1, Features: In step S2, the original haze image is converted from the RGB color space to the HIS color space. The conversion process of the three color channels of red R, green G and blue B in the RGB color space and the three components of hue H, saturation S and brightness I in the HIS color space is as follows: Based on the characteristics of the haze image, the saturation component S of the HIS color space is adjusted, and the brightness component I of the HIS color space is adjusted; the adjusted saturation S O The calculation process of (x, y) is as follows: Where S(x, y) is the initial saturation, S min , S max are the minimum and maximum values ​​of the saturation component respectively; Get the adjusted brightness I V The calculation process of (x, y) is as follows: I V (x,y)=0.4520·I R (x,y)+0.5121·I G (x,y)+0.0359·I B (x,y) In the formula, I R (x, y), I G (x, y), I B (x, y) are the values ​​of the three color components R, G, and B of the haze image respectively.

4. The haze image clearing algorithm based on the atmospheric scattering model according to claim 1, Features: In step S2, the channel brightness difference Z of the haze image κ (x, y) is defined as: The haze image converted to the HIS color space is enhanced by gamma transformation to obtain the brightness value of the haze image after gamma enhancement δ(x, y), and then normalized to obtain the normalized brightness value of the haze image. The calculation process is as follows: Where γ is the gamma parameter, δ min and δ max are the minimum and maximum brightness values ​​of the haze image after gamma enhancement; Construct the distribution probability density function of the dense haze area in the haze image Use the maximum entropy threshold segmentation method to select the segmentation threshold F T ,Based on the distribution probability density function of the dense haze area of ​​the haze image, the haze image is segmented to obtain the dense haze area and the non-dense haze area.

5. The haze image clearing algorithm based on the atmospheric scattering model according to claim 1, Features: In step S3, the haze image is divided into four sub-block areas, and the score of each area is calculated. Determine whether the sub-block area with the highest score belongs to the dense haze area; In the formula, To calculate the average grayscale of each area, n is the number of segmentation, i = 1 represents a dense haze area, and i = 2 represents a non-dense haze area; When the area with the highest score belongs to the non-dense haze area, the segmented area is output; when the area with the highest score belongs to the dense haze area, the area with the highest score is weighted by the distribution probability density function of the dense haze area, and the segmented area is output; The segmented area is divided into four sub-block areas again, and the scores of the re-segmented areas are calculated, and it is determined whether the area with the highest score belongs to the dense haze area; the cycle is repeated until the segmentation termination condition is reached, the segmentation is terminated, and the final segmented area X is output. over ; The last segmentation area X over The first 10% brightness pixels are taken as the global light intensity M(x, y); the segmentation termination condition is: In the formula, is the maximum grayscale mean of the area with the highest segmentation score for the nth time, It is the second largest grayscale mean of the area with the highest segmentation score for the nth time.

6. The haze image clearing algorithm based on the atmospheric scattering model according to claim 5, Features: When the two final segmented regions are obtained, the scores of the last two regions are calculated based on the average gray value and the average gradient value. And compare; the last segmentation area with the highest score is taken as the last segmentation area X over ; In the formula, Represents the average grayscale value of the region. Represents the average gradient of the region.

7. The haze image clearing algorithm based on the atmospheric scattering model according to claim 1, Features: In step S4, based on the principle of sunlight physical model, the atmospheric scattering coefficient of the dense haze area is obtained, and the grayscale information of the illumination map is integrated to obtain the atmospheric scattering coefficient β of the dense haze area. 1 (x, y): In the formula, and τ are fixed coefficients, g(x, y) is the grayscale value of the illumination map corresponding to the dense haze area; A linear model is established and the gradient information of the dense haze image area is integrated to calculate the scene depth of the dense haze area: d 1 (x, y) = a 1 I(x,y)+α 2 S(x,y)+α 3 G(x,y)+ε In the formula, I(x, y), S(x, y), G(x, y) represent the brightness, saturation and gradient information of the image in the dense haze area, respectively, and α 1 , α 2 , α 3 is the linear coefficient, ε is the system random error; The atmospheric scattering coefficient β in the dense haze area 1 (x, y) and the scene depth d of the dense haze area 1 Substituting (x, y) into the transmittance calculation formula, we can get the transmittance t of the dense haze area. 1 (x, y); The transmittance of the dense haze area and the global light intensity are combined to calculate the atmospheric light value A of the dense haze area 1 (x, y).

8. The haze image clearing algorithm based on the atmospheric scattering model according to claim 7, Features: In step S4, the calculation process of the atmospheric light value in the non-dense haze area is as follows; A 2 (x,y)=M(x,y) The transmittance t of the non-dense haze area is obtained by the dark channel prior algorithm 2 (x, y): In the formula, C is any channel of the color channel in RGB, A 2 (x, y) is the atmospheric light value in the non-dense haze area, ω(x, y) represents a window centered at (x, y) in the non-dense haze area, Refers to areas without heavy haze.

Citation Information

Cited By

  • Video stream defogging method and system for 5G remote control

    CN121582108A

  • Haze image synthesis method based on depth guidance and domain alignment

    CN121810506A

  • A haze image synthesis method based on deep guidance and domain alignment

    CN121810506B

  • Defogging adaptive method for synthesizing remote sensing image based on pseudo fog

    CN121937323A

  • Video image rain and fog removal enhancement processing method and system for security and protection monitoring

    CN122223342A