Self-adaptive underwater image enhancement method based on turbidity grading

Abstract

The invention discloses an adaptive underwater image enhancement method based on turbidity grading. The method comprises the following steps: firstly, defining the blurring degree of an image by using an image gradient function, and further dividing an underwater image into a light-turbidity image, a medium-turbidity image and a high-turbidity image; carrying out color correction on the light-turbidity image and then carrying out contrast enhancement; after color correction is carried out on the medium-turbidity image, the medium-turbidity image is converted into a Lab color space, defogging operation and multi-scale detail enhancement are carried out in an L channel, color saturation restoration is carried out on a channel a and a channel b, and finally, a multi-scale detail enhancement image and a color saturation restoration image are fused; a high-turbidity image is firstly filtered, then color correction is carried out, the image is converted to a Lab color space, defogging operation is carried out in an L channel, and finally fusion processing is carried out. According to the method, the turbidity of the water body is graded by using the fuzzy function, and different enhancement strategies are adopted for images with different turbidity, so that high-quality clear images are obtained.

Description

Technical Field

[0001] This invention relates to image enhancement methods, and more particularly to an adaptive underwater image enhancement method based on turbidity classification. Background Technology

[0002] The ocean is a vast treasure trove of resources. With the rapid development of science and technology, human activities underwater are becoming increasingly frequent, encompassing multiple fields such as seabed resource exploration, underwater archaeology, aquaculture, and marine military operations. These activities primarily rely on detection equipment to acquire marine information, and visible light images, as one of the important sources of information for underwater exploration, directly affect the effectiveness of marine scientific research and engineering applications due to their superior realism and richness of information. However, light is scattered by suspended particles in the water during its underwater propagation, leading to reduced visibility in underwater optical images, a phenomenon particularly pronounced in turbid media. Furthermore, different wavelengths of light attenuate at different rates in water, causing color deviations in underwater images. These problems severely restrict the application and development of images in underwater missions.

[0003] Currently, there is a wealth of research on underwater image enhancement, and many effective methods exist. However, these methods and studies generally suffer from a common problem: most are designed for single underwater environments. Natural aquatic environments are not static; water bodies often change gradually with time, temperature, weather, and human influence. Therefore, single underwater image enhancement methods struggle to cope with complex and variable aquatic environments, and their processing results are often unstable due to turbidity variations, failing to meet the general requirements of high-quality imaging. In view of this, this invention proposes an adaptive underwater image enhancement method based on turbidity classification. This method identifies the turbidity level of the water body using an image sharpness index and adaptively matches the optimal enhancement algorithm for different levels, thereby overcoming the limitations of single algorithms and significantly improving the enhancement effect and robustness in various real underwater environments. Summary of the Invention

[0004] This invention provides an adaptive underwater image enhancement method based on water turbidity classification, with the aim of (1) achieving turbidity classification of water bodies and (2) adaptive image enhancement for different aquatic environments. This invention primarily addresses the problems of blurriness, color distortion, and low contrast in underwater images by addressing these issues in stages. It classifies different water bodies by turbidity and then adopts different enhancement strategies based on the characteristics of turbidity in different water bodies, thereby achieving effective underwater image enhancement.

[0005] The technical means employed in this invention are as follows:

[0006] An adaptive underwater image enhancement method based on water turbidity classification includes the following steps:

[0007] S1. Acquire raw underwater images and perform turbidity classification on them;

[0008] S2. Statistically determine the turbidity classification threshold by analyzing the sharpness evaluation values ​​of the Turbid dataset;

[0009] S3. Classify turbidity according to the turbidity threshold, and then use different enhancement algorithms for different turbidity levels;

[0010] Step S3 includes:

[0011] S31. For images with slight turbidity, perform contrast enhancement after color correction;

[0012] S32. After color correction, the moderately turbid image is converted to the Lab color space. Dehazing and multi-scale detail enhancement are performed in the L channel, and color saturation is restored in the a and b channels. Finally, the multi-scale detail enhancement image and the color saturation restoration image are merged.

[0013] S33. For highly turbid images, first perform filtering, then color correction, then convert the image to Lab color space, and perform dehazing in the L channel, and finally perform fusion processing.

[0014] Furthermore, the turbidity classification method in step S1 uses the Tenengrad function to distinguish turbidity, and its calculation formula is as follows:

[0015]

[0016]

[0017] Where T is a given threshold for edge detection, and G x G y It is the convolution of the gradient values ​​in the horizontal and vertical directions at a given pixel.

[0018] Gradient mean of the whole image and the gradient of each sub-region and G grid (m,n), the calculation formula is as follows:

[0019]

[0020]

[0021] in, G represents the overall gradient mean, and G represents the overall image gradient sum. G represents the gradient mean of the subregion. grid (m,n) represents the gradient sum of the sub-region, M×N is the size of the entire image, and m×n is the size of the sub-region;

[0022] The sharpness weighting factor β(m,n) based on the sub-region is obtained by the following formula:

[0023] ;

[0024] The improved algorithm's image gradient formula is as follows:

[0025] ;

[0026] The final improved sharpness rating is obtained by the following formula:

[0027] .

[0028] Furthermore, step S2, based on the sharpness assessment value calculated in step S1, the turbidity classification after setting the sharpness threshold can be expressed by the following formula:

[0029]

[0030] Where, x 1， x2 is a threshold determined after testing with the underwater image dataset Turbid, with values ​​of 10 and 30.

[0031] Furthermore, in step S31, for underwater images with slight turbidity, after color correction, the CLAHE algorithm is further used to enhance local contrast, thereby effectively enhancing key features such as edges and textures.

[0032] Furthermore, in step S32, for the underwater image with moderate turbidity, after color correction, it is converted to the Lab color space, and dehazing is performed in the L channel. The dehazing process is to perform weighted guided image filtering on the L channel.

[0033] Assuming the guiding image is I j The input image is P i The filtered output image is q i Its local linear model expression is:

[0034]

[0035] In the formula: a k b k These are the linear coefficients that remain unchanged within the window, and their values ​​are determined by minimizing the cost function, which is expressed as:

[0036]

[0037] In the formula, λ is a regularization parameter greater than zero, λa k 2 It's a penalty item; avoid excessively large 'a' values.k ;

[0038] A weighted guided filter is proposed, employing an edge weight factor based on a window pixel method. To adjust the penalty coefficient, the above formula becomes:

[0039]

[0040] Using a method similar to guided image filtering, the filtered output image would be:

[0041]

[0042] in and a k and b k The average value;

[0043] The fog image X(p) is decomposed into a base layer and a detail layer using WGIF, with both the input and guide images being the fog image X(p):

[0044]

[0045] Where X b and X e These are the base layer and detail layer of the fog image, respectively. The noise and detail information of the image are located in the detail layer. (Refer to the guided image filtering (GIF) a) k and b k Solving for the base layer X of the fog image X(p). b for:

[0046]

[0047] in and a k and b k Average value within the filter window;

[0048] S322. Based on the dark channel prior theory, in the image base layer X... b The initial transmission rate t of the fog image is calculated as follows:

[0049]

[0050] in, It is a window centered at pixel p with a radius of ρ1, p ’ For window Ω ρ1 Pixels within (p);

[0051] The dehazed image is then represented as:

[0052]

[0053] A represents global atmospheric light;

[0054] After image dehazing, the contrast is not high. A multi-scale detail enhancement method is used to further improve image details. The specific formula is as follows:

[0055]

[0056]

[0057] G1, G2, and G3 are three Gaussian kernels with standard deviations of 1.0, 2.0, and 3.0, respectively. After filtering the image, three underwater images with different degrees of blur are obtained: B1, B2, and B3.

[0058] Finally, the three detail images are linearly summed with appropriate weights to reduce artifact generation. The linear summation formula is as follows:

[0059] ;

[0060] To further enhance the color, a light scattering model is used to compensate for the color during underwater light propagation. The compensation formula is as follows:

[0061]

[0062]

[0063] Where θ represents a constant, t represents the transmission rate, a and b represent the color components of the color spaces, and corr represents the color component of the color space. a corr b This represents the color components of the a and b color spaces after compensation.

[0064] Finally, the multi-scale detail enhancement image D and the color saturation compensated corr are compared. a corr b Channels are fused to generate a high-quality image. The fusion formula is as follows:

[0065]

[0066] Among them, I men Lab represents the fused and enhanced image;

[0067] Convert the image from Lab color space to RGB color space.

[0068] Furthermore, in step S33, for images of highly turbid water bodies, homomorphic filtering is first performed to remove noise. The specific formula is as follows: The image is represented using an illumination-reflection model.

[0069]

[0070] Where f(x,y) represents the grayscale image of the original underwater image, i(x,y) represents the incident component, r(x,y) represents the reflected component, and (x,y) represents the image coordinates;

[0071] The Fourier transform of the above equation is:

[0072]

[0073] H(u,v) is the transfer function of the homomorphic filter;

[0074] Performing an inverse Fourier transform on the frequency-domain filtered image and taking its inverse value to restore it to the spatial domain, the formula is as follows:

[0075]

[0076]

[0077] in , These represent the incident component and the illumination component of the image after filtering, respectively.

[0078] The filtered image is color-corrected, then converted to the Lab color space, and dehazing is performed on the L channel. The dehazing process is as follows:

[0079] Based on the dark channel prior theory, the initial transmission rate t of the fog image in the L channel is calculated as follows:

[0080]

[0081] Among them, Ω ρ1 (p) is a window centered at pixel p with a radius of ρ1, where p ’ For window Ω ρ1 Pixels within (p);

[0082] The dehazed image is then represented as:

[0083] ;

[0084] Finally, the dehazed image and the color-saturation compensated a and b channels are fused to generate a high-quality image. The fusion formula is as follows:

[0085]

[0086] Among them, I hen Lab represents the fused and enhanced image;

[0087] Convert the image from Lab color space to RGB color space;

[0088] Histogram normalization is used to stretch the RGB channels of an image to the range of 0-255, thereby enhancing the grayscale range of the image and improving its global contrast. The histogram normalization algorithm is as follows:

[0089]

[0090] in , The image is the histogram normalized. The image after dehazing. and These are the pixel values ​​at the cumulative histogram values ​​of 0.1% and 99.9%, respectively. The stretched image has a wider histogram distribution, improving overall image contrast.

[0091] The effect achieved by this invention is that the method described in this invention uses fuzzy functions to classify the turbidity of water bodies, adopts different enhancement strategies for images of different turbidities, and obtains high-quality, clear images, thus achieving outstanding technical results. Attached Figure Description

[0092] Figure 1 A flowchart illustrating an adaptive underwater image enhancement method based on turbidity classification, provided in an embodiment of the present invention, is as follows:

[0093] Figure 2 This is a comparison image showing the color correction and contrast enhancement of underwater turbid images by this invention and other algorithms.

[0094] Figure 3 This serves as an objective evaluation metric after processing by this invention and other algorithms. Detailed Implementation

[0095] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0096] See appendix Figure 1 This invention provides an adaptive underwater image enhancement method based on water turbidity classification, comprising the following steps:

[0097] S1. Acquire raw underwater images and perform turbidity classification to obtain clarity assessment values.

[0098] S11. In this step, after acquiring the original underwater image, the acquired image is classified for turbidity. The Tenengrad function is used to distinguish the turbidity, and its calculation formula is as follows:

[0099]

[0100]

[0101] Where T is a given threshold for edge detection, and G x G y It is the convolution of the gradient values ​​in the horizontal and vertical directions at a given pixel.

[0102] S12. Furthermore, to improve the accuracy of turbidity recognition and reduce the impact of sparse image content regions on the overall image evaluation, assuming the image size is M×N, the image is decomposed into K sub-regions with a resolution of m×n, each sub-region being the same size. The average gradient of the entire image is then calculated. and the gradient of each sub-region and G grid (m,n), the calculation formula is as follows:

[0103]

[0104]

[0105] in, G represents the overall gradient mean, and G represents the overall image gradient sum. G represents the gradient mean of the subregion. grid (m,n) represents the gradient sum of the subregion.

[0106] S13. The sharpness weighting factor β(m,n) based on the sub-region can be obtained from the following formula:

[0107]

[0108] S14. The improved algorithm's image gradient formula is as follows:

[0109]

[0110] S15. Finally, the improved sharpness evaluation value is obtained, which is derived from the following formula:

[0111] .

[0112] S21. Based on the sharpness assessment value calculated in step S1, the turbidity classification after setting the sharpness threshold can be expressed by the following formula:

[0113]

[0114] Where, x 1， x2 is a threshold determined after testing with the underwater image dataset Turbid, with values ​​of 10 and 30.

[0115] S31. For images with slight turbidity, perform color correction followed by contrast enhancement.

[0116] In this step, for underwater images with slight turbidity, color correction is performed, and the CLAHE algorithm is further used to enhance local contrast, thereby effectively enhancing key features such as edges and textures.

[0117] S32. For moderately turbid images, after color correction, convert them to the Lab color space. Perform dehazing and multi-scale detail enhancement on the L channel, and restore color saturation on the a and b channels. Finally, merge the multi-scale detail enhanced image and the color saturation restored image. This includes the following steps:

[0118] S321. For underwater images with moderate turbidity, after color correction, convert them to the Lab color space and perform dehazing in the L channel. The dehazing process involves weighted guided image filtering on the L channel.

[0119] Assuming the guiding image is I j The input image is P i The filtered output image is q i Its local linear model expression is:

[0120]

[0121] In the formula: a k b k These are the linear coefficients that remain unchanged within the window, and their values ​​are determined by minimizing the cost function, which is expressed as:

[0122]

[0123] In the formula, λ is a regularization parameter greater than zero, λa k 2 It's a penalty item; avoid excessively large 'a' values. k .

[0124] A weighted guided filter is proposed, employing an edge weight factor based on a window pixel method. To adjust the penalty coefficient, the above formula becomes:

[0125]

[0126] Using a similar approach to Guided Image Filtering (GIF), the filtered output image would be:

[0127]

[0128] in and a k and b k The average value.

[0129] This invention utilizes WGIF to decompose a fog image X(p) into a base layer and a detail layer, where both the input image and the guide image are fog images X(p):

[0130]

[0131] Where X b and X e These are the base layer and detail layer of the fog image, respectively. The noise and detail information of the image are located in the detail layer, and the reference image filter is a. k and b k Solving for the base layer X of the fog image X(p). b for:

[0132]

[0133] in and a k and b k Average value within the filter window.

[0134] S322. Based on the dark channel prior theory, in the image base layer X... b The initial transmission rate t of the fog image is calculated as follows:

[0135]

[0136] in, It is a window centered at pixel p with a radius of ρ1, p ’ For window Ω ρ1 The pixels within (p).

[0137] The dehazed image is then represented as:

[0138]

[0139] S323. After image dehazing, the contrast is not high. A multi-scale detail enhancement method is used to further improve image details. The specific formula is as follows:

[0140]

[0141]

[0142] G1, G2, and G3 are three Gaussian kernels with standard deviations of 1.0, 2.0, and 3.0, respectively. After filtering the image, three underwater images with different degrees of blur are obtained: B1, B2, and B3.

[0143] Finally, the three detail images are linearly summed with appropriate weights to reduce artifact generation. The linear summation formula is as follows:

[0144]

[0145] S324. To further enhance the color, a light scattering model is used to compensate for the color during underwater light propagation. The compensation formula is as follows:

[0146]

[0147]

[0148] Where θ represents a constant, t represents the transmission rate, a and b represent the color components of the color spaces, and corr represents the color component of the color space. a corr b This represents the color components of the a and b color spaces after compensation.

[0149] S325. Finally, the multi-scale detail enhancement image D and the color saturation compensated corr are compared. a corr b Channels are fused to generate a high-quality image. The fusion formula is as follows:

[0150]

[0151] Among them, I men Lab represents the fused enhanced image, and Lab is the fusion formula.

[0152] S326. Convert the image from Lab color space to RGB color space;

[0153] S33. For highly turbid images, filtering is performed first, followed by color correction. Then, the image is converted to the Lab color space, and dehazing is performed in the L channel. Finally, a fusion process is applied. This includes the following steps:

[0154] S331. In image processing of highly turbid water bodies, homomorphic filtering is first performed to remove noise. The specific formula is as follows: The image is represented using an illumination-reflection model.

[0155]

[0156] Where f(x,y) represents the grayscale image of the original underwater image, i(x,y) represents the incident component, r(x,y) represents the reflected component, and (x,y) represents the image coordinates.

[0157] Further performing a Fourier transform on the above equation yields:

[0158]

[0159] H(u,v) is the transfer function of the homomorphic filter.

[0160] Further performing an inverse Fourier transform on the frequency domain filtered image and taking the inverse value to restore it to the spatial domain, the formula is as follows:

[0161]

[0162]

[0163] in , These represent the incident component and the illumination component of the image after filtering, respectively.

[0164] S332. Perform color correction on the filtered image, then convert it to the Lab color space, and perform dehazing on the L channel. The dehazing process is as follows:

[0165] The defogging process is as follows:

[0166] Based on the dark channel prior theory, the initial transmission rate t of the fog image in the L channel is calculated as follows:

[0167]

[0168] Among them, Ω ρ1 (p) is a window centered at pixel p with a radius of ρ1, where p ’ For window Ω ρ1 The pixels within (p).

[0169] The dehazed image is then represented as:

[0170]

[0171] S333. Finally, the dehazed image and the color-saturation compensated a and b channels are fused to generate a high-quality image. The fusion formula is:

[0172]

[0173] Among them, I hen Lab represents the fused enhanced image, and Lab is the fusion formula.

[0174] S334. Convert the image from Lab color space to RGB color space;

[0175] S335. Histogram normalization is used to stretch the RGB channels of the image to the range of 0-255, thereby enhancing the grayscale range of the image and improving the global contrast. The histogram normalization algorithm is as follows:

[0176]

[0177] in , The image is the histogram normalized. The image after dehazing. and These are the pixel values ​​at the cumulative histogram values ​​of 0.1% and 99.9%, respectively. After stretching, the histogram range of the image is larger, which improves the overall contrast of the image.

[0178] Example

[0179] like Figure 2 The image shown illustrates the experimental results of this invention compared to other algorithms in processing heavily turbid underwater images. The results clearly demonstrate that after turbidity grading and color correction using this invention, the image colors are more realistic, and both global and local contrast are significantly enhanced. In contrast, methods such as UDCP, ACDC, CBAF, CDCR, IBLA, MLLE, PCDE, PUIE-Net, and TACL show poor color correction and detail enhancement effects and cannot adapt to different turbidity levels. Therefore, the method of this invention can effectively correct image colors while improving image detail.

[0180] Figure 3 The objective evaluation results of different methods provided in the embodiments of the present invention are compared.

[0181] In the table, upward arrows (↑) indicate that higher values ​​result in better performance, while downward arrows (↓) indicate the opposite. It can be seen that this invention achieved the best results among the six indicators and ranked second best in the remaining indicator. The table shows that the proposed method exhibits a high overall evaluation value, indicating that it has a more significant effect on enhancing image quality compared to other methods.

[0182] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An adaptive underwater image enhancement method based on turbidity grading, characterized in that, The method includes: S1. Acquire raw underwater images and perform turbidity classification to obtain clarity assessment values; S2. Statistically determine the turbidity classification threshold by analyzing the sharpness evaluation values ​​of the Turbid dataset; S3. Classify turbidity according to the turbidity threshold, and then use different enhancement algorithms for different turbidity levels; Step S3 includes: S31. For images with slight turbidity, perform contrast enhancement after color correction; S32. After color correction, the moderately turbid image is converted to the Lab color space. Dehazing and multi-scale detail enhancement are performed in the L channel, and color saturation is restored in the a and b channels. Finally, the multi-scale detail enhancement image and the color saturation restoration image are merged. S33. For highly turbid images, first perform filtering, then color correction, then convert the image to Lab color space, and perform dehazing in the L channel, and finally perform fusion processing.

2. The adaptive underwater image enhancement method based on turbidity grading according to claim 1, characterized in that, The turbidity classification method in step S1 uses the Tenengrad function to distinguish turbidity, and its calculation formula is as follows: Where T is a given threshold for edge detection, and G x G y It is the convolution of the gradient values ​​in the horizontal and vertical directions at a given pixel. Gradient mean of the whole image and the gradient of each sub-region and G grid (m,n), the calculation formula is as follows: in, G represents the overall gradient mean, and G represents the overall image gradient sum. G represents the gradient mean of the subregion. grid (m,n) represents the gradient sum of the sub-region, M×N is the size of the entire image, and m×n is the size of the sub-region; The sharpness weighting factor β(m,n) based on the sub-region is obtained by the following formula: ； The improved algorithm's image gradient formula is as follows: ； The final improved sharpness rating is obtained by the following formula: 。 3. The adaptive underwater image enhancement method based on turbidity grading according to claim 1, characterized in that, Step S2 involves calculating sharpness evaluation values ​​for all images in the Turbid dataset using the sharpness evaluation values ​​from step S1, statistically analyzing the turbidity classification threshold, and finally determining the turbidity classification threshold. The turbidity classification after setting the sharpness threshold is expressed by the following formula: Where, x 1， x2 is a threshold determined after testing with the underwater image dataset Turbid, with values ​​of 10 and 30.

4. The adaptive underwater image enhancement method based on turbidity grading according to claim 1, characterized in that, In step S31: For underwater images with slight turbidity, color correction is performed, and the CLAHE algorithm is further used to enhance local contrast, thereby effectively enhancing key features of edges and textures.

5. The adaptive underwater image enhancement method based on turbidity grading according to claim 1, characterized in that, In step S32: For underwater images with moderate turbidity, after color correction, they are converted to the Lab color space, and dehazing is performed in the L channel. The dehazing process involves weighted guided image filtering of the L channel. Assuming the guiding image is I j The input image is P i The filtered output image is q i Its local linear model expression is: In the formula: a k b k These are the linear coefficients that remain unchanged within the window, and their values ​​are determined by minimizing the cost function, which is expressed as: In the formula, λ is a regularization parameter greater than zero, λa k 2 It's a penalty item; avoid excessively large 'a' values. k ; A weighted guided filter is proposed, employing an edge weight factor based on a window pixel method. To adjust the penalty coefficient, the above formula becomes: Using a method similar to guided image filtering, the filtered output image would be: in and a k and b k The average value; The fog image X(p) is decomposed into a base layer and a detail layer using WGIF, with both the input and guide images being the fog image X(p): Where X b and X e These are the base layer and detail layer of the fog image, respectively. The noise and detail information of the image are located in the detail layer, and the reference image filter is a. k and b k Solving for the base layer X of the fog image X(p). b for: in and a k and b k Average value within the filter window; Based on the dark channel prior theory, in the image base layer X b The initial transmission rate t of the fog image is calculated as follows: in, It is a window centered at pixel p with a radius of ρ1, p ’ For window Ω ρ1 Pixels within (p); The dehazed image is then represented as: A represents global atmospheric light; After image dehazing, the contrast is not high. A multi-scale detail enhancement method is used to further improve image details. The specific formula is as follows: G1, G2, and G3 are three Gaussian kernels with standard deviations of 1.0, 2.0, and 3.0, respectively. After filtering the image, three underwater images with different degrees of blur are obtained: B1, B2, and B3. Finally, the three detail images are linearly summed with appropriate weights to reduce artifact generation. The linear summation formula is as follows: ； To further enhance the color, a light scattering model is used to compensate for the color during underwater light propagation. The compensation formula is as follows: Where θ represents a constant, t represents the transmission rate, a and b represent the color components of the color spaces, and corr represents the color component of the color space. a corr b This represents the color components of the a and b color spaces after compensation. Finally, the multi-scale detail enhancement image D and the color saturation compensated corr are compared. a corr b Channels are fused to generate a high-quality image. The fusion formula is as follows: Among them, I men Lab represents the fused and enhanced image; Convert the image from Lab color space to RGB color space.

6. The adaptive underwater image enhancement method based on turbidity grading according to claim 1, characterized in that, In step S33: For images of highly turbid water bodies, homomorphic filtering is first performed to remove noise. The specific formula is as follows: The image is represented using an illumination-reflection model. Where f(x,y) represents the grayscale image of the original underwater image, i(x,y) represents the incident component, r(x,y) represents the reflected component, and (x,y) represents the image coordinates; The Fourier transform of the above equation is: H(u,v) is the transfer function of the homomorphic filter; Performing an inverse Fourier transform on the frequency-domain filtered image and taking its inverse value to restore it to the spatial domain, the formula is as follows: in , These represent the incident component and the illumination component of the image after filtering, respectively. The filtered image is color-corrected, then converted to the Lab color space, and dehazing is performed on the L channel. The dehazing process is as follows: Based on the dark channel prior theory, the initial transmission rate t of the fog image in the L channel is calculated as follows: Among them, Ω ρ1 (p) is a window centered at pixel p with a radius of ρ1, where p ’ For window Ω ρ1 Pixels within (p); The dehazed image is then represented as: ； Finally, the dehazed image and the color-saturation compensated a and b channels are fused to generate a high-quality image. The fusion formula is as follows: Among them, I hen Lab represents the fused and enhanced image; Convert the image from Lab color space to RGB color space; Histogram normalization is used to stretch the RGB channels of an image to the range of 0-255, thereby enhancing the grayscale range of the image and improving its global contrast. The histogram normalization algorithm is as follows: in , The image is the histogram normalized. The image after dehazing. and These are the pixel values ​​at the cumulative 0.1% and 99.9% of the histogram, respectively.

Images