Image restoration method based on adaptive color correction, back scattering light optimization and scene transmittance improvement

Abstract

The invention relates to an image restoration method based on adaptive color correction, back scattering light optimization and scene transmittance improvement, which comprises the following steps: calculating a total value and an information retention factor of each color channel, and selecting a reference channel; calculating an adjustment coefficient; compensating a low pixel value; determining whether the maximum color loss is greater than a loss threshold; dividing the color correction image into a structure component and a texture component; estimating a scene transmittance on the structure component; estimating backscattered light on the structure component; and obtaining a clear image according to the enhanced structure component and the enhanced texture component. According to the method, adaptive color correction is realized, contrast enhancement is optimized, robustness is realized, and more detailed information is provided while the natural appearance of the whole image is kept; the maximum saturation constraint is adopted for scene transmission rate estimation, block artifacts can be avoided under ideal conditions, the edge of a significant object is enhanced, white noise and misjudgment of the object are eliminated, and a more accurate result can be generated.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision, and in particular to an image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance. Background Art

[0002] Currently, visual images have been widely used in target tracking, ocean resource development, underwater vehicles, etc. However, due to the interference of attenuation effects, visual images usually have problems such as color distortion, low contrast, and blurred details. These degradation problems hinder the application performance of visual images in many important explorations. Therefore, it is crucial to propose a simple and effective image restoration method.

[0003] Generally speaking, methods for enhancing degraded images can be divided into three categories: model-based methods, model-free methods, and deep learning-based methods. Among them, model-based methods usually construct a physical model of imaging by considering the degradation process of degraded images and use handcrafted priors to invert the physical model to generate clear images. These commonly used priors include dark channel prior, super Laplacian reflectance prior, adaptive dark pixel prior, illumination channel sparse prior, etc. However, complex scenes have different lighting conditions and attenuation characteristics, making these methods often unstable in image restoration.

[0004] Model-free methods often enhance the details, colors, and contrast of degraded images by directly adjusting the pixel values or histogram distributions in the images. Such methods, such as histogram equalization methods, image fusion-based methods, etc., can improve the image quality to a certain extent, but are prone to over-enhancement or under-enhancement enhancement results.

[0005] Deep learning-based methods utilize the powerful learning ability of neural networks on the basis of large-scale image data to improve the overall contrast and color of underwater images. However, deep learning-based methods such as convolutional neural network-based methods usually require a large amount of paired image data for network training and are difficult to apply in dynamic remote sensing scenes.

[0006] Therefore, how to design a more effective image restoration method has become an urgent technical problem to be solved. Summary of the Invention

[0007] To solve the problems of artifacts, poor visibility, and over-enhancement or under-enhancement that usually exist in existing underwater image enhancement methods for complex illumination images, the purpose of the present invention is to provide an image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance that can avoid block artifacts, enhance the edges of significant objects, and provide more detailed information while retaining the natural appearance of the entire image.

[0008] To achieve the above object, the present invention adopts the following technical solutions: An image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance, the method comprising the following steps in sequence:

[0009] (1) Calculate the total value of each color channel and the information retention factor Select the maximum information retention factor The corresponding channel is used as the reference channel I Ref ;

[0010] (2) According to the maximum information retention factor and the information retention factors corresponding to the channels that need color compensation, calculate the adjustment coefficient ω of each color channel c ; The two channels other than the reference channel are the channels that need color compensation;

[0011] (3) Compensate for low pixel values through the adjustment coefficient ω of the channels that need color compensation c and the reference channel I Ref to obtain a color-corrected image

[0012] (4) Calculate the maximum color loss Loss of the color-corrected image max , and determine whether the maximum color loss Loss max is greater than the loss threshold of 0.02; if the judgment result is yes, repeat steps (1) to (4) until the maximum color loss Loss max is less than or equal to the loss threshold of 0.02; if the judgment result is no, proceed to the next step;

[0013] (5) Use the Gaussian filter G to divide the color-corrected image into a structure component and a texture component

[0014] (6) Use the adjusted saturation to estimate the scene transmittance t on the structure component c ;

[0015] (7) Construct a scoring formula to estimate the backscattered light A on the structure component c ;

[0016] (8) Calculate the enhanced structure component and the enhanced texture component, and obtain the clear image J c (x, y).

[0017] Step (1) specifically refers to: for the underwater image I c , calculate the total value of each color channel

[0018]

[0019] where H and W respectively represent the rows and columns of the input underwater image I c ; r, g, and b respectively represent the red, green, and blue channels of the input underwater image I c in the RGB space; I c (x, y) is the pixel value of each color channel at the pixel coordinates (x, y); c is the color channel of the underwater image I c ;

[0020] Use the total value I of each color channel c t to calculate the information retention factor of each color channel

[0021]

[0022] where respectively represent the total values of the red, green, and blue channels;

[0023] Select the channel corresponding to the maximum information retention factor as the reference channel I Ref , The formula for is:

[0024]

[0025] where max() represents the function used to construct the maximum value, respectively represent the information retention factors of the red, green, and blue channels.

[0026] In step (2), the calculation formula for ω c is:

[0027]

[0028] Step (3) specifically refers to: the color correction image The formula for is:

[0029]

[0030] where (x, y) are the pixel coordinates; I c (x, y) is the pixel value of each color channel of the underwater image at the pixel coordinates (x, y); I Ref (x, y) is the reference channel I RefThe pixel value at the pixel coordinates (x, y).

[0031] In step (4), the maximum color loss Loss max has the formula:

[0032]

[0033] where are the average values of the red, green, and blue channels of the image respectively.

[0034] Step (5) specifically refers to:

[0035] The formula for the imaging model of the underwater scene is:

[0036] I c (x,y) = J c (x,y)t(x,y) + A(1 - t(x,y)) (1)

[0037] where I c (x,y) represents the pixel value of each color channel of the underwater image at the pixel coordinates (x, y), t(x,y) is the scene transmittance, and A is the backscattered light;

[0038] From formula (1), it can be seen that to obtain J c (x,y), it is necessary to estimate the scene transmittance t(x,y) and the backscattered light A;

[0039] The color-corrected image is divided into a structure component and a texture component. The structure component contains the overall structure and contour of the underwater image, and the texture component shows details and textures; the structure component is generated based on the Gaussian filter G and the texture component

[0040]

[0041] where σ represents the standard deviation;

[0042] Perform a convolution operation on the input I c (x,y) and the Gaussian filter G to generate the structure component

[0043]

[0044] Then use the difference between I c (x,y) and the structure component to obtain the texture component

[0045]

[0046] Step (6) specifically refers to:

[0047] According to the imaging model of the underwater scene, the structural component is written as:

[0048]

[0049] where represents the clear image on the structural component;

[0050] The clear image on the structural component of the dark channel is:

[0051]

[0052] where Ω(x,y) represents the local window centered at (x,y);

[0053] Two minimum filters are used to estimate the scene transmission rate t on the structural component c :

[0054]

[0055] where represents the estimated scene transmission rate on the structural component ;

[0056] Formula (4) is derived as the following formula:

[0057]

[0058] Substituting formula (3) into formula (5) and further deriving the following formula:

[0059]

[0060] According to the definition of the dark channel, the dark channel in formula (6) is expressed as:

[0061]

[0062] where:

[0063]

[0064] represents the minimum value of the structural component I S (u,v) in the red, green, and blue channels; represents the maximum value of the structural component I S (u,v) in the red, green, and blue channels;

[0065] Meanwhile, the brightness V(x, y) and saturation S(x, y) in the HSV color space on the structural component are respectively expressed as:

[0066]

[0067] After substituting formulas (8) and (9) into formula (7), formula (7) is transformed into:

[0068]

[0069] Combining formula (6) and formula (10), the scene transmittance t c is approximately expressed as:

[0070]

[0071] If it is assumed that the local brightness of the underwater image is constant, formula (11) is written as:

[0072]

[0073] Among them, the saturation of each pixel satisfies the inequality:

[0074]

[0075] Two methods are used to adjust the saturation S(x, y):

[0076] The first method uses S(x, y) to estimate the scene transmittance, and formula (12) is written as:

[0077]

[0078] The second method is to enhance the image saturation based on the maximum saturation constraint, and calculate the difference d(x, y) between the saturation value of each pixel and the upper limit threshold S max :

[0079] d(x, y) = min((S max - S(x, y)) 2 , (1 - |S max - S(x, y)|) 2 )(14)

[0080] On the premise of formula (14), the following equation is used to obtain the average value Υ of the difference d(x, y) between the saturation value of each pixel and the upper limit threshold S max :

[0081]

[0082] Among them, MN is the size of the input image;

[0083] The scale of saturation adjustment depends on the difference degree α between the saturation value of each pixel and the upper threshold S max :

[0084]

[0085] where is a control coefficient for controlling the adjustment range, and the formula is:

[0086]

[0087] The updated adjusted saturation S new (x, y) is expressed as:

[0088]

[0089] Estimate the scene transmittance t through the adjusted saturation connection formula (13) c .

[0090] Step (7) specifically refers to: using the brightness of the image as one of the criteria for locating the backscattering region, where the calculation formula of the image brightness is as follows:

[0091]

[0092] where h and w are the height and width of the sub-region respectively, and W k,brightness represents the brightness of the k-th sub-region in the image, represents the pixel value of the k-th sub-region in the color correction image at the pixel point coordinates (x, y); using the standard deviation as a criterion for identifying the backscattering region, where the calculation formula of the standard deviation is as follows:

[0093]

[0094] where W k,contrast represents the standard deviation of the k-th sub-region in the image, represents the average value of the k-th sub-region;

[0095] Add an effective attenuation difference constraint for the backscattering region to reconstruct the objective function, where the attenuation difference W of the k-th sub-region backscattering region k,diff is:

[0096]

[0097] Therefore, combining equations (15) to (17), the scoring formula for the backscattering region is defined as:

[0098]

[0099] The specific steps for locating the backscattering region are as follows:

[0100] (7a) Divide the input image into four rectangular blocks;

[0101] (7b) Calculate the score value for each rectangular block;

[0102] (7c) Take the rectangular block with the largest score value among the four blocks as the backscattering candidate region;

[0103] (7d) Then continue to repeat steps (7a) to (7c) in the backscattering candidate region until the size of the rectangular block with the largest score value is less than 32*32, and take the rectangular block with a size less than 32*32 as the backscattering region; finally, select the brightest pixel value in the backscattering region as the backscattering light A c .

[0104] Step (8) specifically refers to: Restoring the clear image on the structural component according to the scene transmittance t c Restoring the clear image on the structural component

[0105]

[0106] Using gamma correction to enhance the illumination to obtain the enhanced structural component

[0107]

[0108] where τ = 0.6;

[0109] The relationship between the gradient and the scene transmittance is inversely proportional on the image, which is expressed by the following formula:

[0110]

[0111] where, represents the gradient operator; t represents the normalized scene transmittance, and I c represents the input underwater image;

[0112] Applying the reciprocal of the scene transmittance estimated on the structural component to enhance the gradient on the texture component:

[0113]

[0114] where ω is a non - negative coefficient used to control the enhancement scale of gradient restoration; represents the gradient of the original texture component of the underwater image; represents the gradient of the enhanced texture component of the underwater image; Represents the average value of the scene transmittance; with the help of the enhanced gradient and Poisson equation, an enhanced texture component is generated ; where Restore refers to the gradient restoration operation

[0115]

[0116] Finally, using the enhanced structure component

[0117] and the enhanced texture component a clear image J is obtained c (x, y):

[0118]

[0119] As can be seen from the above technical solutions, the beneficial effects of the present invention are as follows: First, the present invention realizes adaptive color correction, optimizes contrast enhancement, and has robustness, providing more detailed information while retaining the natural appearance of the entire image; Second, the adaptive color correction strategy in the present invention utilizes the principle of maximum color loss to design a compensation factor for the channel that needs color compensation to obtain a uniform pixel distribution; Third, the present invention uses the maximum saturation constraint for scene transmission rate estimation, which can avoid block artifacts and enhance the edges of significant objects under ideal conditions; Fourth, the present invention constructs a new scoring formula to estimate the backscattered light, which can eliminate white noise and misjudgment of objects and can produce more accurate results BRIEF DESCRIPTION OF THE DRAWINGS

[0120] Figure 1 is the flowchart of the method of the present invention

[0121] Figure 2 is the visual comparison result diagram DETAILED DESCRIPTION OF THE EMBODIMENTS

[0122] As Figure 1 shown, an image restoration method based on adaptive color correction, optimizing backscattered light, and improving scene transmittance, the method includes the following steps in sequence

[0123] (1) Calculate the total value of each color channel and the information retention factor Select the channel corresponding to the maximum information retention factor as the reference channel I Ref ;

[0124] (2) Calculate the adjustment coefficient ω of each color channel according to the maximum information retention factor and the information retention factor corresponding to the channel that needs color compensation c; The two channels other than the reference channel are the channels that require color compensation;

[0125] (3) Compensate for the low pixel values through the adjustment coefficient ω c of the channels that require color compensation Ref and the reference channel I

[0126] (4) Calculate the maximum color loss Loss of the color-corrected image max , and determine whether the maximum color loss Loss max is greater than the loss threshold of 0.02; if the judgment result is yes, repeat steps (1) to (4) until the maximum color loss Loss max is less than or equal to the loss threshold of 0.02; if the judgment result is no, proceed to the next step; According to the gray world assumption: in natural images, the average value and histogram distribution of each color channel are almost the same. Therefore, in order to conform to this assumption, iterate and repeat steps (1) to (4) until the maximum color loss Loss max is less than or equal to the loss threshold of 0.02.

[0127] (5) Use the Gaussian filter G to divide the color-corrected image into a structure component and a texture component

[0128] (6) Use the adjusted saturation to estimate the scene transmittance t on the structure component c ;

[0129] (7) Construct a scoring formula to estimate the backscattered light A on the structure component c ;

[0130] (8) Calculate the enhanced structure component and the enhanced texture component, and obtain the clear image J c (x, y).

[0131] Step (1) specifically refers to: For the underwater image I c , calculate the total value of each color channel

[0132]

[0133] where H and W respectively represent the rows and columns of the input underwater image I c ; r, g, b respectively represent the red, green, and blue channels of the input underwater image I c in the RGB space; Ic (x, y) is the pixel value of each color channel at the pixel coordinates (x, y); c is the color channel of the underwater image I c ;

[0134] Using the total value of each color channel Calculate the information retention factor of each color channel

[0135]

[0136] Among them, respectively represent the total values of the red, green, and blue channels;

[0137] Select the channel corresponding to the maximum information retention factor as the reference channel I Ref , The formula of is:

[0138]

[0139] Among them, max() represents the function used to construct the maximum value, respectively represent the information retention factors of the red, green, and blue channels.

[0140] In step (2), the formula for ω c is:

[0141]

[0142] Step (3) specifically refers to: the color correction image The formula of is:

[0143]

[0144] Among them, (x, y) are the pixel coordinates; I c (x, y) is the pixel value of each color channel of the underwater image at the pixel coordinates (x, y); I Ref (x, y) is the pixel value of the reference channel I Ref at the pixel coordinates (x, y).

[0145] In step (4), the formula for the maximum color loss Loss max is:

[0146]

[0147] Among them, are respectively the averages of the red, green, and blue channels of the image.

[0148] Step (5) specifically refers to:

[0149] The formula for the imaging model of the underwater scene is as follows:

[0150] I c (x, y) = J c (x, y)t(x, y) + A(1 - t(x, y))(1)

[0151] Where, I c (x, y) represents the pixel value of each color channel of the underwater image at the pixel point coordinates (x, y), t(x, y) is the scene transmittance, and A is the backscattered light;

[0152] It can be seen from formula (1) that to obtain J c (x, y), it is necessary to estimate the scene transmittance t(x, y) and the backscattered light A;

[0153] Due to the influence of suspended particles, underwater images often have a large amount of high-frequency textures and noises. These factors will interfere with the estimation of the scene transmittance and the backscattered light, thus affecting the restoration performance. To solve these limitations, the color-corrected image is divided into a structure component and a texture component. The structure component contains the overall structure and contour of the underwater image, and the texture component shows details and textures; therefore, to avoid the interference of high-frequency textures and noises, the scene transmittance t c (x, y) and the backscattered light A c are estimated on the structure component. The structure component is generated based on the Gaussian filter G and the texture component

[0154]

[0155] Where, σ represents the standard deviation;

[0156] Perform a convolution operation on the input I c (x, y) and the Gaussian filter G to generate the structure component

[0157]

[0158] After that, use the difference between I c (x, y) and the structure component to obtain the texture component

[0159]

[0160] Step (6) specifically refers to:

[0161] According to the imaging model of the underwater scene, the structure component is written as:

[0162]

[0163] Among them, represents the clear image on the structural component;

[0164] It is known that the method for estimating the scene transmittance based on DCP assumes that in a clear image, the pixel value in at least one local small patch on one channel tends to zero. Based on this, the clear image on the structural component of the dark channel is:

[0165]

[0166] where Ω(x,y) represents the local window centered at (x,y);

[0167] Two minimum filters are used to estimate the scene transmission rate t on the structural component c :

[0168]

[0169] Among them, represents the estimated scene transmission rate on the structural component ;

[0170] Formula (4) is derived as the following formula:

[0171]

[0172] Substituting formula (3) into formula (5) and further deriving the following formula:

[0173]

[0174] According to the definition of the dark channel, the dark channel in formula (6) is expressed as:

[0175]

[0176] where:

[0177]

[0178] represents the minimum value of the structural component I S (u,v) in the red, green, and blue channels; represents the maximum value of the structural component I S (u,v) in the red, green, and blue channels;

[0179] Meanwhile, the brightness V(x, y) and saturation S(x, y) in the HSV color space on the structural component are respectively expressed as:

[0180]

[0181] After substituting formulas (8) and (9) into formula (7), formula (7) is transformed into:

[0182]

[0183] It can be seen from formulas (6) and (10) that the estimated scene transmittance on the structural component is highly correlated with the saturation and brightness on the structural component. Meanwhile, for most underwater images, the value of the backscattered light is almost the same and relatively large. Combining formulas (6) and (10), the scene transmittance t c is approximately expressed as:

[0184]

[0185] If it is assumed that the local brightness of the underwater image is constant, formula (11) is written as:

[0186]

[0187] where the saturation of each pixel satisfies the inequality:

[0188]

[0189] From this condition, it can be known that formula (12) will overestimate the scene transmittance, that is, the larger the patch size, the larger the scene transmittance value. In addition, formula (6) is smaller than formula (5) in a real and complex underwater environment, and formula (12) is derived from formula (6), which may lead to an underestimation of the scene transmittance. To solve the above problems, two methods are used to adjust the saturation S(x, y):

[0190] The first method uses S(x, y) to estimate the scene transmittance, and formula (12) is written as:

[0191]

[0192] The second method is to enhance the image saturation based on the maximum saturation constraint, and calculate the difference d(x, y) between the saturation value of each pixel and the upper threshold S max :

[0193] d(x, y) = min((S max - S(x, y)) 2 , (1 - |S max - S(x, y)|) 2 )(14)

[0194] On the premise of formula (14), the average value Υ of the difference d(x, y) between the saturation value of each pixel and the upper threshold S is obtained using the following equation: max between:

[0195]

[0196] where MN is the size of the input image;

[0197] The scale of saturation adjustment depends on the degree of difference α between the saturation value of each pixel and the upper threshold S max :

[0198]

[0199] where represents the control coefficient for controlling the adjustment range, and the formula is:

[0200]

[0201] The updated adjusted saturation S new (x, y) is expressed as:

[0202]

[0203] The scene transmittance t is estimated through the adjusted saturation connection formula (13) c .

[0204] Step (7) specifically refers to: using the prior of the backscattering region to construct a scoring rule to estimate the backscattered light A c . It is known that the brightness of the backscattering region is usually very high. Therefore, using the brightness of the image becomes one of the criteria for locating the backscattering region. The calculation formula for the image brightness is as follows:

[0205]

[0206] where h and w are the height and width of the sub-region respectively, and W k,brightness represents the brightness of the k-th sub-region in the image, represents the pixel value of the k-th sub-region in the color-corrected image at the pixel point coordinates (x, y);

[0207] In addition, because the backscattering region is almost contributed by the backscattered light A c , that is, the contrast of the backscattered light region is low, and the brightness difference between different positions is not large. Therefore, the standard deviation is used as a criterion for identifying the backscattering region. The calculation formula for the standard deviation is as follows:

[0208]

[0209] Among them, W k,contrast represents the standard deviation of the k-th sub-region in the image, and represents the average value of the k-th sub-region;

[0210] Add a valid backward scattering region attenuation difference constraint to reconstruct the objective function. Among them, the attenuation difference W of the backward scattering region of the k-th sub-region k,diff is:

[0211]

[0212] Therefore, combining equations (15) to (17), the scoring formula for the backward scattering region is defined as:

[0213]

[0214] The specific steps for locating the backward scattering region are as follows:

[0215] (7a) Divide the input image into four rectangular blocks;

[0216] (7b) Calculate the score for each rectangular block;

[0217] (7c) Take the rectangular block with the largest score among the four blocks as the backward scattering candidate region;

[0218] (7d) Then continue to repeat steps (7a) to (7c) in the backward scattering candidate region until the size of the rectangular block with the largest score is less than 32*32, and take the rectangular block with a size less than 32*32 as the backward scattering region; finally, select the brightest pixel value in the backward scattering region as the backward scattering light A c .

[0219] Step (8) specifically refers to: Restoring the clear image on the structural component according to the scene transmittance t c Since water significantly absorbs the energy of the incident light, the image illuminance on the structural component will be relatively low after removing the backward scattering light in formula (18). To solve this problem, gamma correction is used to enhance the illumination to obtain the enhanced structural component

[0220]

[0221]

[0222]

[0223] ​Among them, τ = 0.6. The present invention can remove the haze on the structural component and improve the contrast. However, the details on the structural component are still insufficient, and the problem can be solved by enhancing the details on the texture component.

[0224] The relationship between the gradient and the scene transmittance is inversely proportional on the image, which is expressed by the following formula:

[0225]

[0226] Among them, represents the gradient operator; t represents the normalized scene transmittance, and I c represents the input underwater image;

[0227] Apply the reciprocal of the scene transmittance estimated on the structural component to enhance the gradient on the texture component:

[0228]

[0229] Among them, ω is a non - negative coefficient used to control the enhancement scale of gradient restoration; represents the gradient of the original texture component of the underwater image; represents the gradient of the enhanced texture component of the underwater image; represents the average value of the scene transmittance ; With the help of the enhanced gradient and the Poisson equation, an enhanced texture component

[0230]

[0231] Among them, Restore refers to the gradient restoration operation;

[0232] Finally, use the enhanced structural component and the enhanced texture component to obtain the clear image J c (x, y):

[0233]

[0234] In Figure 2 , the original image of the underwater image input in the embodiment of the present invention is as shown in Figure 2 (a), the underwater image processed by the prior - art LANet is as shown in Figure 2 (b), the underwater image processed by the prior - art PUIENet is as shown in Figure 2 (c), the underwater image processed by the prior - art DeepWav is as shown in Figure 2 (d), the underwater image processed by the prior - art HAAM is as shown in Figure 2As shown in (e), the underwater image processed by the prior art LEPFNet is as Figure 2 As shown in (f), the underwater image processed by the prior art LiteNet is as Figure 2 As shown in (g), the underwater image processed by the prior art ShallowNet is as Figure 2 As shown in (h), the underwater image processed by the prior art DAUT is as Figure 2 As shown in (i), the underwater image processed by the prior art UShape is as Figure 2 As shown in (j), the underwater image processed by the prior art PCDE is as Figure 2 As shown in (k), the underwater image processed by the present invention is as Figure 2 As shown in (l). It can be seen that the underwater image processing result of the present invention has natural and true color saturation, rich details and sufficient brightness.

[0235] In summary, the present invention realizes adaptive color correction, optimizes contrast enhancement, and has robustness, providing more detailed information while retaining the natural appearance of the whole image; the adaptive color correction strategy in the present invention utilizes the principle of maximum color loss and designs a compensation factor corresponding to the channel that needs color compensation to obtain a uniform pixel distribution; the present invention uses the maximum saturation constraint to estimate the scene transmission rate, which can avoid block artifacts and enhance the edges of significant objects in an ideal situation; fourth, the present invention constructs a new scoring formula to estimate the backscattered light, which can eliminate white noise and misjudgment of objects and can produce more accurate results.

Claims

1. An image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance, characterized in that: The method includes the following steps in sequence: (1) Calculate the total value of each color channel and the information retention factor Select the maximum information retention factor The corresponding channel is used as the reference channel I Ref ; (2)According to the maximum information retention factor and the information retention factor corresponding to the channel that requires color compensation, calculate the adjustment coefficient ω of each color channel c ; The two channels other than the reference channel are the channels that require color compensation; (3) The adjustment coefficient ω of the channel that requires color compensation c and the reference channel I Ref are used to compensate for the low pixel values to obtain a color-corrected image (4) Calculate the color-corrected image of the maximum color loss Loss max , and determine whether the maximum color loss Loss max is greater than the loss threshold of 0.02; if the judgment result is yes, repeat steps (1) to (4) until the maximum color loss Loss max is less than or equal to the loss threshold of 0.02; if the judgment result is no, proceed to the next step; (5) Divide the color-corrected image using the Gaussian filter G into a structural component and a texture component (6) Estimate the structure component using the adjusted saturation of the scene transmittance t c : (7) Construct a scoring formula to estimate the structural component Backscattered light A on c ; (8) Calculate the enhanced structure component and the enhanced texture component, and obtain the clear image J c (x, y).

2. The image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance according to claim 1, wherein: Step (1) specifically refers to: for the underwater image I c , calculate the total value of each color channel where H and W respectively represent the rows and columns of the input underwater image I c ; r, g, and b respectively represent the red, green, and blue channels of the input underwater image I c in the RGB color space; I c (x, y) is the pixel value of each color channel at the pixel coordinates (x, y); c is the color channel of the underwater image I c ; Using the total value I of each color channel c t Calculate the information retention factor of each color channel Among them, respectively represent the total values of the red, green, and blue channels; Select the maximum information retention factor The corresponding channel is used as the reference channel I Ref , The formula is as follows: where max() represents a function for constructing the maximum value, respectively representing the information retention factors of the red, green, and blue channels.

3. The image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance according to claim 1, characterized in that: In step (2), the ω c is calculated by the following formula:

4. The image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance according to claim 1, characterized in that: Step (3) specifically refers to: the color correction image The formula for: where (x, y) are the pixel coordinates; I c (x, y) are the pixel values of each color channel of the underwater image at the pixel coordinates (x, y); I Ref (x, y) is the reference channel I Ref at the pixel values at the pixel coordinates (x, y).

5. The image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance according to claim 1, characterized in that: In step (4), the maximum color loss Loss max is calculated by the formula: Among them, are the average values of the red, green, and blue channels of the image, respectively.

6. The image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance according to claim 1, wherein: Step (5) specifically refers to: The formula of the imaging model for the underwater scene is: I c (x, y) = J c (x, y)t(x, y) + A(1 - t(x, y))(1) Among them, I c (x, y) represents the pixel value of each color channel of the underwater image at the pixel coordinates (x, y), t(x, y) is the scene transmittance, and A is the backscattered light; As can be seen from Equation (1), to obtain J c (x, y), it is necessary to estimate the scene transmittance t(x, y) and the backscattered light A; Divide the color-corrected image into a structure component and a texture component. The structure component contains the overall structure and contour of the underwater image, and the texture component shows details and textures; generate the structure component I based on the Gaussian filter G c S and the texture component Where σ represents the standard deviation; Convolve the input I c (x, y) with the Gaussian filter G to generate the structure component After using I c (x, y) and the structural component The difference between them results in the texture component 7. The image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance according to claim 1, characterized in that: Step (6) specifically refers to: According to the imaging model of the underwater scene, the structure component is written as: Among them, represents a clear image on the structural component; Clear image on the structural component Dark channel is as follows: Where Ω(x, y) represents the local window centered at (x, y); Use two minimum filters to estimate the structure component of the scene transmission rate t c : Among them, represents the scene transmission rate estimated on the structural component ​ Formula (4) is derived into the following formula: Substituting formula (3) into formula (5) and further deriving the following formula: According to the definition of the dark channel, in formula (6), the dark channel is expressed as: Where: Represents the structural component I S (u, v) is the minimum value in the red, green, and blue channels; Represents the structural component I S (u, v) is the maximum value in the red, green, and blue channels; At the same time, the brightness V(x, y) and saturation S(x, y) in the HSV color space of the structural component are respectively expressed as: After substituting formula (8) and formula (9) into formula (7), formula (7) is transformed into: Combining formula (6) and formula (10), the scene transmittance t c is approximately expressed as: If it is assumed that the local brightness of the underwater image is constant, formula (11) is written as: Where the saturation of each pixel satisfies the inequality: Adjust the saturation S(x, y) using two methods: The first method uses S(x, y) to estimate the scene transmittance, and formula (12) is written as: The second method is to enhance the image saturation based on the maximum saturation constraint, and calculate the difference d(x, y) between the saturation value of each pixel and the upper threshold S max : d(x,y) = min((S max - S(x,y)) 2 , (1 - |S max - S(x,y)|) 2 )(14) On the premise of formula (14), the average value Υ of the difference d(x, y) between the saturation value of each pixel and the upper threshold S is obtained using the following equation: max between: Where MN is the size of the input image; The scale of saturation adjustment depends on the degree of difference α between the saturation value of each pixel and the upper threshold S max : Among them, represents a control coefficient for controlling the adjustment range, and the formula is: Updated and adjusted saturation S new (x, y) is expressed as: Estimate the scene transmittance t through the adjusted saturation connection formula (13) c .

8. The image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance according to claim 1, wherein: Step (7) specifically refers to: using the brightness of the image as one of the criteria for locating the backscattering region, and the calculation formula of the image brightness is as follows: where h and w are the height and width of the sub-region, respectively, and W k,brightness represents the brightness of the k-th sub-region in the image, represents the pixel value of the k-th sub-region in the color-corrected image at the pixel coordinates (x, y); the standard deviation is used as a criterion for identifying the backscattering region, and the calculation formula for the standard deviation is as follows: Among them, W k,contrast represents the standard deviation of the k-th sub-region in the image, and represents the average value of the k-th sub-region; Add a valid backscattering region attenuation difference constraint to reconstruct the objective function, where the attenuation difference W of the backscattering region of the k-th sub-region k,diff is as follows: Therefore, combining equations (15) to (17), the scoring formula for the backscattering region is defined as: The specific steps for locating the backscattering region are: (7a) Divide the input image into four rectangular blocks; (7b) Calculate the score value for each rectangular block; (7c) Use the rectangular block with the largest score among the four blocks as the backscattering candidate region; (7d) Then continue to repeat steps (7a) to (7c) in the backscattering candidate region until the size of the rectangular block with the maximum score is less than 32*32, and use this rectangular block with a size less than 32*32 as the backscattering region; finally, select the brightest pixel value in the backscattering region as the backscattered light A c .

9. The image restoration method based on adaptive color correction, optimized backscattered light, and improved scene transmittance according to claim 1, wherein: Step (8) specifically refers to: according to the scene transmittance t c Restore the clear image on the structural component Gamma correction is used to enhance the illumination to obtain an enhanced structural component Where τ = 0.6; The relationship between the gradient and the scene transmittance is inversely proportional on the image, and is expressed by the following formula: Among them, represents the gradient operator; t represents the normalized scene transmittance, and I c represents the input underwater image; Apply the reciprocal of the scene transmittance estimated on the structural component to enhance the gradient on the texture component: where ω is a non - negative coefficient used to control the enhancement scale of gradient recovery; represents the gradient of the original texture component of the underwater image; represents the gradient of the enhanced texture component of the underwater image; represents the average value of the scene transmittance ; with the help of the enhanced gradient and Poisson equation, an enhanced texture component is generated Where Restore refers to the gradient restoration operation; Finally, using the enhanced structure component and the enhanced texture component obtain the clear image J c (x, y):

Images