Underwater low-illumination image enhancement method based on local color correction and multi-scale variational decomposition model

Abstract

The invention relates to an underwater low-illumination image enhancement method based on local color correction and a multi-scale variation decomposition model, and belongs to the technical field of image processing. The method comprises the following steps: firstly, performing color correction on an input underwater low-illumination image; next, the corrected image is converted from an RGB color space to an HSV color space, a multi-scale variational decomposition model is applied to a V channel in the image, and a base layer, two detail layers and an atomization layer are generated; then, the atomization layer is separated, self-adaptive brightness equalization is carried out on the base layer, an enhanced base layer is obtained, self-adaptive detail sharpening is carried out on the detail layer, and an enhanced detail layer is obtained; and finally, linearly fusing the enhanced base layer and detail layer to obtain an optimized V channel, converting the image back to an RGB color space, and outputting an enhanced underwater image. According to the method, the overall enhancement of the image is realized, the contrast, the brightness and the detail expression are remarkably optimized, and the visual effect of the underwater image is remarkably improved.

Description

Technical Field

[0001] The invention relates to an underwater low-illuminance image enhancement method based on local color correction and a multi-scale variational decomposition model, and belongs to the technical field of image processing. Background Art

[0002] In recent years, with the increasing depletion of land resources, the development and utilization of marine resources and the protection of the marine environment have gradually become the focus of global attention, and underwater detection technology has also become a key supporting technology and research hotspot in this field. Underwater optical imaging technology has gradually become the preferred means of close-range underwater detection due to its advantages such as fast imaging speed and high resolution. However, underwater images are easily affected by many factors during natural light imaging: first, the selective absorption of light by water causes serious color shift in the image; second, the scattering characteristics of water make the image blurred and produce a fogging effect. These factors usually lead to color distortion, blurred details and reduced contrast in underwater images, which increases the difficulty of image analysis and understanding. In addition, with the increase of depth of field, the absorption effect of water on light becomes more significant, causing the energy of natural light to gradually decay, which leads to the widespread problem of low light or weak light in underwater images. Under the influence of these interference factors, the quality of underwater images is seriously degraded. Existing low-light underwater image enhancement methods often lead to excessive enhancement of bright areas, resulting in obvious color distortion and overexposure. Therefore, developing an underwater low-light image enhancement method that can effectively avoid excessive enhancement of bright areas, color distortion and overexposure has important application value in the field of underwater image processing. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide an underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model, aiming to solve the technical problem that existing low-light underwater image enhancement methods often lead to excessive enhancement of bright areas, resulting in obvious color distortion and overexposure, making it difficult to ensure the visual quality of underwater images.

[0004] To achieve the above objectives, the present invention proposes an underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model. The method corrects the color distortion problem through local color compensation technology, and decomposes the V channel based on the multi-scale variational decomposition model to obtain the base layer, detail layer and fog layer; the fog layer is extracted to enhance the contrast; the base layer is subjected to an adaptive brightness equalization method to enhance the brightness of the dark area and maintain the brightness of the normal bright area; at the same time, the detail layer is subjected to adaptive detail sharpening to restore clear texture information.

[0005] The technical solution of the present invention is: an underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model, comprising the following steps:

[0006] Step 1: Input an underwater low-light image, and apply local color compensation technology to the three channels of R (red), G (green) and B (blue) of the underwater low-light image respectively to achieve adaptive color correction and generate a color-compensated image;

[0007] Step 2: Convert the color-compensated image to the HSV color space and separate it into three channels: H (hue), S (saturation), and V (value). Apply a multi-scale variational decomposition model to the V channel to decompose it into a base layer, two detail layers, and a fog layer.

[0008] Step 3: Perform adaptive brightness equalization on the base layer, adaptive sharpening on the detail layer, and separation on the fog layer;

[0009] Specifically, an adaptive brightness equalization method is used for the base layer to optimize the image brightness and obtain an enhanced base layer; an adaptive detail sharpening process is performed on the detail layer to improve the clarity of texture information and obtain an enhanced detail layer; and a fog layer is extracted to enhance the image contrast;

[0010] Step 4: Combine the processed base layer and detail layer through linear fusion to generate an optimized V channel, and then convert the image converted to the HSV color space back to the RGB color space to obtain the enhanced underwater image.

[0011] The specific method of applying the local color compensation technology to the three channels R, G and B of the underwater low-light image in step 1 is as follows:

[0012] Step 1.1: Decompose the input underwater low-light image into local blocks of size M×N, and calculate the average value of the R, G, and B channels in each local block respectively. The specific calculation formula is as follows:

[0013]

[0014] Among them, I r (x), I g (x), I b (x) represents the R, G, and B channels of the input image respectively; m r 、m g 、m b Respectively represent the average values ​​of the R, G, and B channels of the input image; H and W are the height and width of the input channel;

[0015] Step 1.2: From the average values ​​of the R, G, and B channels calculated in the local block, select the maximum value as the reference channel. The specific calculation formula is as follows:

[0016]

[0017] Among them, I ref (x) indicates the reference channel;

[0018] Step 1.3: Calculate the color compensation factor corresponding to each channel according to the calculated average values ​​of the R, G and B channels in the local block. The specific calculation formula is as follows:

[0019]

[0020] Among them, z r 、z g 、z b Respectively represent the color compensation factors of R, G, and B channels; m ref is the reference channel I ref The average value of (x);

[0021] Step 1.4: Substitute the color compensation factors corresponding to each channel into the R, G and B channels in the corresponding local block to obtain the image after color compensation. The calculation formula is as follows:

[0022]

[0023] in, Represents a color-corrected image.

[0024] The specific method of applying the multi-scale variational decomposition model to the V channel in step 2 is as follows:

[0025] Step 2.1: Model the V channel through a mathematical model, which is specifically composed as follows:

[0026] V=B1+D1+ΔH (9)

[0027] Among them, B1 represents the base layer, which contains rich structural information and brightness information; D1 represents the detail layer, which contains texture information; ΔH represents the fog layer after being processed by the guided filter;

[0028] Step 2.2: Use the multi-scale variational decomposition model to process the V channel, and the objective function is defined as follows:

[0029]

[0030] Among them, ||·||2, ||·||1 and ||·||0 represent the Euclidean norm, the one-norm and the zero-norm respectively; ▽ represents the first-order gradient operator; represents the dot multiplication operation of image pixels; λ1, λ2, δ correspond to the constant parameters of each regularization term, G B and G D are the weight matrices representing the base layer and detail layer, respectively, which can be expressed by the following mathematical formula:

[0031]

[0032] Among them, G B (B1(i,j)) represents the weight matrix corresponding to the pixel B1(i,j) of the base layer B1; G D (D1(i,j)) represents the weight matrix corresponding to the pixel D1(i,j) of the detail layer D1; M represents the intermediate operation; f is the maximum neighborhood difference model; Ω represents a local area of ​​size 3×3;

[0033] Step 2.3: Use the alternating direction multiplier method to solve the base layer B1, detail layer D1 and fog layer ΔH, and introduce two auxiliary variables X and C into the objective function, so that the objective function is redefined as:

[0034]

[0035] Introducing two Lagrangian function multipliers Z1 and Z2 removes the equality constraints of the objective function, and the objective function can be represented by the Lagrangian function It is expressed as:

[0036]

[0037] Where <·,·> represents the inner product of the matrix; μ is the penalty scalar; and the Lagrangian function It can be solved by iterating to update each variable and treating other irrelevant variables as constants;

[0038] For the base layer B1, ignoring other irrelevant variables, the corresponding objective function is expressed as:

[0039]

[0040] Where k is the number of iterations; represents the base layer after the k+1th iteration update; ΔH k , X k , μ k Respectively represent the detail layer D1, the fog layer ΔH, the Lagrangian function multiplier Z1, the auxiliary variable X, and the penalty scalar μ after the k-th iteration update; ▽ represents the first-order gradient operator;

[0041] The base layer after the k+1th iteration is updated using the two-dimensional fast Fourier transform technique. Perform the calculation:

[0042]

[0043] in, represents a two-dimensional fast Fourier transform operation, represents a two-dimensional inverse Fourier transform operation; for The complex conjugate of Represents the dot multiplication operation of image pixels; W d is the matrix from the discrete gradient operator, containing the horizontal gradient h and vertical gradient v directions;

[0044] For the detail layer D1, ignoring other irrelevant variables, the corresponding objective function is expressed as:

[0045]

[0046] The updated detail layer D1 after the k+1th iteration k+1 The solution is as follows:

[0047]

[0048] For the atomization layer ΔH, after fixing other variables except ΔH, the corresponding objective function is expressed as:

[0049]

[0050] After obtaining the updated base layer and detail layer after the k+1th iteration, the updated ΔH after the k+1th iteration k+1 It is obtained by solving the closed form solution of the subproblem. The specific calculation formula is as follows:

[0051]

[0052] For the X component, the related items with X are used to form an energy optimization function to solve the sub-problem of the X component, which is expressed as follows:

[0053]

[0054] The closed-form solution of formula (21) is achieved through the contraction operation, so the X component after the k+1th iteration update is expressed as follows:

[0055]

[0056] Among them, X k+1 is the X component after the k+1th iteration update, For the convergence operation, it is expressed by the following mathematical formula:

[0057] S t (x)=sign(x)·max(x|-t,0) (23)

[0058] Among them, sign(·) is the sign function; max(·) is the maximization function;

[0059] For the C component, after collecting the items about the C component, the corresponding objective function is expressed as:

[0060]

[0061] Formula (24) is solved element by element, so the C component after the k+1th iteration update is expressed by a piecewise function:

[0062]

[0063] Among them, C k+1 is the C component after the k+1th iteration update;

[0064] For the Lagrange multipliers Z1 and Z2 corresponding to the k+1th iteration updated Lagrange multipliers and And the penalty scalar μ corresponding to the penalty scalar μ after the k+1th iteration update k+1 , the specific iteration process is defined as follows:

[0065]

[0066] μ k+1 =4·μ k (28)

[0067] when and or and The error between -3 Or when the total number of iterations k exceeds the preset threshold K (k=15), the entire iteration process stops;

[0068] The base layer B1 is decomposed multiple times to eliminate artifacts, and the following variational model is used:

[0069]

[0070] Among them, B2 represents the second base layer, which is the base layer obtained by the final decomposition; λ3 is a constant set by experience; then the second detail layer is expressed as:

[0071] D2=B1-B2 (30)

[0072] Among them, D2 is the second detail layer.

[0073] The specific method of performing adaptive brightness equalization processing on the base layer and adaptive sharpening processing on the detail layer in step 3 is as follows:

[0074] Step 3.1: Extract the base layer B2 of the V channel;

[0075] Step 3.2: Adaptively balance the brightness of the base layer B2. The calculation formula is as follows;

[0076]

[0077] in, is the enhanced base layer; B 2_max and B 2_min denote the maximum and minimum values ​​of the base layer B2, respectively; exp(·) is an exponential function;

[0078] Step 3.3: Adaptively sharpen the detail layers D1 and D2. The calculation formula is as follows:

[0079]

[0080] in, represents the enhanced detail layers D1 and D2; max(·) is the maximization function; sign(·) is the composite function; is the average value of the V channel; D i Represents the detail layers D1, D2, ω i Represents the weight coefficient of the i-th layer, which can be expressed by the following calculation formula:

[0081]

[0082] Among them, ω1 is the weight coefficient of the first detail layer; ω2 is the weight coefficient of the second detail layer.

[0083] In step 4, the processed base layer and detail layer are combined by linear fusion to generate the optimized V channel as follows:

[0084]

[0085] Among them, V E is the optimized V channel; Represents the enhanced base layer; and They are represented as the enhanced first detail layer and second detail layer respectively.

[0086] The present invention has outstanding substantive features and significant progress compared to the prior art. Specifically, the beneficial effects of the present invention are:

[0087] 1. This method can simultaneously solve the color distortion and low contrast problems caused by the absorption and scattering of light in the underwater environment, and effectively deal with the underexposure phenomenon caused by light energy attenuation;

[0088] 2. This method is applicable to a variety of visual application scenarios in deep-sea environments, such as underwater archaeology, deep-sea robots, and other tasks;

[0089] 3. In addition to enhancing underwater low-light images, this method is also applicable to other degraded scenarios, including night imaging, sandstorm weather, and improvement of remote sensing image quality.

[0090] Based on the above reasons, the present invention can be widely promoted in the fields of image processing with color cast degradation. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0092] Figure 1 It is a schematic diagram of the process of the present invention;

[0093] Figure 2 This is the enhancement result of the underwater low-illumination image by the present invention and other underwater image processing methods. DETAILED DESCRIPTION

[0094] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0095] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0096] Example 1: Figure 1 As shown, an underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model is provided, comprising the following steps:

[0097] Step 1: Input an underwater low-light image, and apply local color compensation technology to the three channels of R, G and B of the underwater low-light image respectively to achieve adaptive color correction and generate a color-compensated image;

[0098] Specifically, due to the selective absorption characteristics of light in underwater environments, the attenuation of different spectra varies, which causes color deviation problems. In order to compensate for these attenuation channels and achieve a balance between different color channels, thereby effectively correcting color distortion, a local color compensation technology is proposed. First, the underwater low-light image is divided into local blocks of size 5×5, and the average value of each local block is calculated, and one of them is selected as the reference channel. The specific calculation formula is as follows:

[0099]

[0100] Among them, I r (x), I g (x), I b (x) represents the R, G, and B channels of the input image respectively; m r 、m g 、m b Respectively represent the average values ​​of the R, G, and B channels of the input image; H and W are the height and width of the input channel;

[0101] Next, the maximum value among the average values ​​of R, G, and B channels calculated from the local block is selected as the reference channel. The specific calculation formula is as follows:

[0102]

[0103] Among them, I ref (x) indicates the reference channel;

[0104] According to the calculated average values ​​of R, G and B channels in the local block, the color compensation factor corresponding to each channel is derived. The specific calculation formula is as follows:

[0105]

[0106] Among them, z r 、z g 、z b Respectively represent the color compensation factors of R, G, and B channels; m ref is the reference channel I ref The average value of (x);

[0107] Next, the color compensation factor of each channel is substituted into the corresponding local block R, G and B channels to obtain the image after color compensation. The specific calculation formula of this process is as follows:

[0108]

[0109] in, represents a color-corrected image;

[0110] Step 2: Convert the color-compensated image to the HSV color space and separate it into three channels: H, S, and V. Apply a multi-scale variational decomposition model to the V channel to decompose it into a base layer, two detail layers, and a fog layer.

[0111] Specifically, although the color-corrected image presents more natural colors, it still has problems such as low contrast, low brightness, and blurred details. To further reduce the impact of color interference, this method converts the color-corrected image from the RGB color space to the HSV color space and processes the V channel. Given that the V channel is mainly affected by brightness, details, and fog information, it can be modeled through a mathematical model. The specific structure is as follows:

[0112] V=B1+D1+ΔH (44)

[0113] Among them, B1 represents the base layer, which contains rich structural information and brightness information; D1 represents the detail layer, which contains texture information; ΔH represents the haze layer after being processed by the guided filter, which helps to avoid the loss of details through the smoothing characteristics of the haze information;

[0114] In order to more accurately separate the physical information in the V channel, this method applies a multi-scale variational decomposition model based on the Euclidean norm, one norm and zero norm to the V channel, and its objective function is defined as follows:

[0115]

[0116] Among them, ||·||2, ||·||1 and ||·||0 represent the Euclidean norm, the one-norm and the zero-norm respectively; ▽ represents the first-order gradient operator; represents the dot multiplication operation of the image pixels; λ1, λ2, and δ correspond to the constant parameters of each regularization term, and in this embodiment, the constant parameters λ1, λ2, and δ are set to 0.3, 0.03, and 0.01, respectively, based on experience. B and G D is the weight matrix, which can be expressed by the following mathematical formula:

[0117]

[0118] Among them, G B (B1(i,j)) represents the weight matrix corresponding to the pixel B1(i,j) of the base layer B1; G D (D1(i,j)) represents the weight matrix corresponding to the pixel D1(i,j) of the detail layer D1; M represents the intermediate operation; f is the maximum neighborhood difference model; Ω represents a local area of ​​size 3×3;

[0119] Specifically, the first term in the above objective function is the data fidelity term, which is used to constrain the solution (B1+D1+ΔH) to be consistent with the brightness channel image V; the second regularization term Used to adjust the piecewise smoothness of the base layer; the third regularization term Used to preserve the continuity of detail layers and salient structures; the fourth regularization term Highlight the smoothness characteristics of the atomized information and constrain the gradient sparsity of the atomized information.

[0120] In order to solve the above objective function, the present invention adopts the alternating direction multiplier method to solve the base layer B1, the detail layer D1 and the haze layer ΔH, and introduces two auxiliary variables X and C into the objective function, so that the objective function can be redefined as:

[0121]

[0122] Then, two Lagrangian function multipliers Z1 and Z2 are introduced to remove the equality constraints of the objective function. Then the objective function can be represented by the Lagrangian function It is expressed as:

[0123]

[0124] Where <·,·> represents the inner product of the matrix; μ is the penalty scalar; and the Lagrangian function It can be solved by iterating to update each variable and treating other irrelevant variables as constants.

[0125] For the base layer B1 subproblem: By eliminating variables that are not related to B1, the corresponding function can be expressed as follows:

[0126]

[0127] Where k is the number of iterations; represents the base layer after the k+1th iteration update; ΔH k , X k , μ k Respectively represent the detail layer D1, the fog layer ΔH, the Lagrangian function multiplier Z1, the auxiliary variable X, and the penalty scalar μ after the k-th iteration update; ▽ represents the first-order gradient operator;

[0128] The above equation is solved by two-dimensional fast Fourier transform technology, and the calculation formula is as follows:

[0129]

[0130] in, represents a two-dimensional fast Fourier transform operation, represents a two-dimensional inverse Fourier transform operation; for The complex conjugate of Represents the dot multiplication operation of image pixels; W d is the matrix from the discrete gradient operator, containing the horizontal gradient h and vertical gradient v directions;

[0131] For the base layer D1 subproblem: By eliminating variables irrelevant to D1, its objective function is defined as follows:

[0132]

[0133] The above equation is solved by two-dimensional fast Fourier transform technology, and the calculation formula is as follows:

[0134]

[0135] in, It is the detail layer after the k+1th iteration update;

[0136] For the subproblem of the fog layer ΔH, by eliminating variables that are irrelevant to ΔH, its objective function is defined as follows:

[0137]

[0138] After obtaining the updated base layer and detail layer after the k+1th iteration, the updated ΔH after the k+1th iteration k+1 It is obtained by solving the closed form solution of the subproblem. The specific calculation formula is as follows:

[0139]

[0140] For the component X subproblem, the optimization function is defined as follows:

[0141]

[0142] Using the convergence operation to solve the component X subproblem can be expressed as follows:

[0143] S t (x)=sign(x)·max(|x|-t,0) (57)

[0144] Among them, sign(·) is the sign function; max(·) is the maximization function; then, the convergence operation is used to solve the X component, and the calculation formula is as follows:

[0145]

[0146] Among them, X k+1 is the X component after the k+1th iteration update, and S is the convergence operation;

[0147] For the sub-problem of component C, the optimization function is defined as follows:

[0148]

[0149] The above function can be solved element by element, so the C component after the k+1th iteration update can be expressed as a piecewise function:

[0150]

[0151] Among them, C k+1 is the C component after the k+1th iteration update;

[0152] Finally, for the Lagrange multipliers Z1 and Z2 corresponding to the k+1th iteration updated Lagrange multipliers and And the penalty scalar μ corresponding to the penalty scalar μ after the k+1th iteration update k+1 , the specific iteration process is defined as follows:

[0153]

[0154] when and or and The error between -3 Or when the total number of iterations k exceeds the preset threshold K (k=15), the entire iteration process stops;

[0155] Although the single-scale decomposition method can provide a constrained base layer, a piecewise constant detail layer, and a smooth fog layer, artifacts may still occur. To this end, the base layer B1 is decomposed multiple times, using the following variational model:

[0156]

[0157] Wherein, B2 represents the second base layer, which is the base layer obtained by the final decomposition; λ3 is an empirically set constant, which is 0.1 in this embodiment; then the second detail layer D2 can be expressed as:

[0158] D2=B1-B2 (65)

[0159] Among them, D2 is the second detail layer;

[0160] Step 3: Perform adaptive brightness equalization on the base layer, adaptive sharpening on the detail layer, and separation on the fog layer;

[0161] Specifically, each layer obtained by decomposition is processed separately: the fog layer is separated to improve the image contrast; the base layer is adaptively brightened to improve the brightness of the dark area and maintain the brightness of the bright area. The calculation formula is as follows:

[0162]

[0163] in, is the enhanced base layer; B 2_max and B 2_min denote the maximum and minimum values ​​of the base layer B2, respectively; exp(·) is an exponential function;

[0164] Adaptive detail sharpening is performed on detail layers D1 and D2, and the calculation formula is as follows:

[0165]

[0166] in, represents the enhanced detail layers D1 and D2; max(·) is the maximization function; sign(·) is the composite function; is the average value of the V channel; D i Represents the detail layers D1, D2, ω i Represents the weight coefficient of the i-th layer, which can be expressed by the following calculation formula:

[0167]

[0168] Among them, ω1 is the weight coefficient of the first detail layer; ω2 is the weight coefficient of the second detail layer;

[0169] Step 4: Combine the processed base layer and detail layer through linear fusion to generate an optimized V channel, and then convert the image converted to the HSV color space back to the RGB color space to obtain the enhanced underwater image.

[0170] Specifically, the enhanced base layer and detail layer are combined by linear fusion to generate an optimized V channel, and the calculation formula is as follows:

[0171]

[0172] Among them, V E is the optimized V channel; Represents the enhanced base layer; and They are represented as the enhanced first detail layer and the second detail layer respectively; finally, the image is converted from the HSV color space back to the RGB color space to obtain the enhanced underwater image.

[0173] In order to verify the advantages of the present invention over the existing methods in processing underwater color-shifted images, a number of underwater low-illuminance images were selected as test sets, and compared with the following underwater image processing methods: Underwater Image Restoration Based on Image Blurriness and Light Absorption (IBLA), Wavelet Pixel Domain Progressive Fusion Network (WPFNET), Hybrid Fusion Method (HFM), Underwater Normalized Total Variation Model (UNTV), Multicolor Components and Light Attenuation (MCLA), Principal Component Fusion of Foreground and Background (PCFB), Texture Enhancement Model Based on Blurriness and Color Fusion (TEBCF), Color-Balanced Locally Adjustable Underwater Image Enhancement Technology (CLEAN-100), and Color-Balanced Locally Adjustable Underwater Image Enhancement Technology (CLEAN-100). ImageEnhancement, CBLA). The comparison results are shown in the form of test images.

[0174] Figure 2 The enhancement results of underwater low-illuminance images by the present invention and other underwater image processing methods are demonstrated. Experiments show that the MCLA and UNTV methods over-enhance the low-illuminance areas, resulting in local overexposure. Although the IBLA, WPENET and CBLA methods can avoid the overexposure problem, there are still obvious color distortion and fogging effects in some areas. In addition, the HFM, PCFB and TEBCF methods over-enhance texture information, resulting in local block effects in the image, thus affecting the visual effect. In contrast, the algorithm of the present invention can not only effectively correct color distortion and enhance image contrast, but also extract more detailed local texture information and achieve brightness balance, significantly improving the overall visual quality of underwater images under low-illuminance conditions.

[0175] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model, characterized in that: The following steps are involved: Step 1: Input an underwater low-light image, and apply local color compensation technology to the three channels of R, G and B of the underwater low-light image respectively to achieve adaptive color correction and generate a color-compensated image; Step 2: Convert the color-compensated image to the HSV color space and separate it into three channels: H, S, and V. Apply a multi-scale variational decomposition model to decompose the V channel into a base layer, two detail layers, and a fog layer. Step 3: Perform adaptive brightness equalization on the base layer, adaptive sharpening on the detail layer, and separation on the fog layer; Step 4: Combine the processed base layer and detail layer by linear fusion to generate an optimized V channel, and then convert the image converted to the HSV color space back to the RGB color space to obtain the enhanced underwater image.

2. The underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model according to claim 1, characterized in that: The specific method of applying the local color compensation technology to the three channels R, G and B of the underwater low-light image in step 1 is as follows: Step 1.1: Decompose the input underwater low-light image into local blocks of size M×N, and calculate the average value of the R, G, and B channels in each local block respectively. The specific calculation formula is as follows: Among them, I r (x), I g (x), I b (x) represents the R, G, and B channels of the input image respectively; m r 、m g 、m b Respectively represent the average values ​​of the R, G, and B channels of the input image; H and W are the height and width of the input channel; Step 1.2: From the average values ​​of the R, G, and B channels calculated in the local block, select the maximum value as the reference channel. The specific calculation formula is as follows: Among them, I ref (x) indicates the reference channel; Step 1.3: Calculate the color compensation factor corresponding to each channel according to the calculated average values ​​of the R, G and B channels in the local block. The specific calculation formula is as follows: Among them, z r 、z g 、z b Respectively represent the color compensation factors of R, G, and B channels; m ref is the reference channel I ref The average value of (x); Step 1.4: Substitute the color compensation factors corresponding to each channel into the R, G and B channels in the corresponding local block to obtain the image after color compensation. The calculation formula is as follows: in, Represents a color-corrected image.

3. The underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model according to claim 1, characterized in that: The specific method of applying the multi-scale variational decomposition model to the V channel in step 2 is as follows: Step 2.1: Model the V channel through a mathematical model, which is specifically composed as follows: V=B1+D1+ΔH (9) Among them, B1 represents the base layer, which contains rich structural information and brightness information; D1 represents the detail layer, which contains texture information; ΔH represents the fog layer after being processed by the guided filter; Step 2.2: Use the multi-scale variational decomposition model to process the V channel, and the objective function is defined as follows: Among them, ||·||2, ||·||1 and ||·||0 represent the Euclidean norm, the one norm and the zero norm respectively; represents the first-order gradient operator; represents the dot multiplication operation of image pixels; λ1, λ2, δ correspond to the constant parameters of each regularization term, G B and G D are the weight matrices representing the base layer and detail layer, respectively, which can be expressed by the following mathematical formula: Among them, G B (B1(i,j)) represents the weight matrix corresponding to the pixel B1(i,j) of the base layer B1; G D (D1(i,j)) represents the weight matrix corresponding to the pixel D1(i,j) of the detail layer D1; M represents the intermediate operation; f is the maximum neighborhood difference model; Ω represents a local area of ​​size 3×3; Step 2.3: Use the alternating direction multiplier method to solve the base layer B1, detail layer D1 and fog layer ΔH, and introduce two auxiliary variables X and C into the objective function, so that the objective function is redefined as: Introducing two Lagrangian function multipliers Z1 and Z2 removes the equality constraints of the objective function, and the objective function can be represented by the Lagrangian function It is expressed as: Where <·,·> represents the inner product of the matrix; μ is the penalty scalar; and the Lagrangian function It can be solved by iterating to update each variable and treating other irrelevant variables as constants; For the base layer B1, ignoring other irrelevant variables, the corresponding objective function is expressed as: Where k is the number of iterations; represents the base layer after the k+1th iteration update; ΔH k , X k , μ k They represent the detail layer D1, the fog layer ΔH, the Lagrangian function multiplier Z1, the auxiliary variable X, and the penalty scalar μ after the k-th iteration update respectively; represents the first-order gradient operator; The base layer after the k+1th iteration is updated using the two-dimensional fast Fourier transform technique. Perform the calculation: in, represents a two-dimensional fast Fourier transform operation, represents a two-dimensional inverse Fourier transform operation; for The complex conjugate of Represents the dot multiplication operation of image pixels; W d is the matrix from the discrete gradient operator, containing the horizontal gradient h and vertical gradient v directions; For the detail layer D1, ignoring other irrelevant variables, the corresponding objective function is expressed as: The updated detail layer after the k+1th iteration The solution is as follows: For the atomization layer ΔH, after fixing other variables except ΔH, the corresponding objective function is expressed as: After obtaining the updated base layer and detail layer after the k+1th iteration, the updated ΔH after the k+1th iteration k+1 It is obtained by solving the closed form solution of the subproblem. The specific calculation formula is as follows: For the X component, the related items with X are used to form an energy optimization function to solve the sub-problem of the X component, which is expressed as follows: The closed-form solution of formula (21) is achieved through the contraction operation, so the X component after the k+1th iteration update is expressed as follows: Among them, X k+1 is the X component after the k+1th iteration update, S is the convergence operation, which is expressed by the following mathematical formula: Among them, sign(·) is the sign function; max(·) is the maximization function; For the C component, after collecting the items about the C component, the corresponding objective function is expressed as: Formula (24) is solved element by element, so the C component after the k+1th iteration update is expressed by a piecewise function: Among them, C k+1 is the C component after the k+1th iteration update; For the Lagrange multipliers Z1 and Z2 corresponding to the k+1th iteration updated Lagrange multipliers and And the penalty scalar μ corresponding to the penalty scalar μ after the k+1th iteration update k+1 , the specific iteration process is defined as follows: when and or and The error between -3 Or when the total number of iterations k exceeds the preset threshold K (k=15), the entire iteration process stops; The base layer B1 is decomposed multiple times to eliminate artifacts, and the following variational model is used: Among them, B2 represents the second base layer, which is the base layer obtained by the final decomposition; λ3 is a constant set by experience; then the second detail layer is expressed as: D2=B1-B2 (30) Among them, D2 is the second detail layer.

4. The underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model according to claim 1, characterized in that: The specific method of performing adaptive brightness equalization processing on the base layer and adaptive sharpening processing on the detail layer in step 3 is as follows: Step 3.1: Extract the base layer B2 of the V channel; Step 3.2: Adaptively balance the brightness of the base layer B2. The calculation formula is as follows; in, is the enhanced base layer; B 2_max and B 2_min denote the maximum and minimum values ​​of the base layer B2, respectively; exp(·) is an exponential function; Step 3.3: Adaptively sharpen the detail layers D1 and D2. The calculation formula is as follows: in, represents the enhanced detail layers D1 and D2; max(·) is the maximization function; sign(·) is the composite function; is the average value of the V channel; D i Represents the detail layers D1, D2, ω i Represents the weight coefficient of the i-th layer, which can be expressed by the following calculation formula: Among them, ω1 is the weight coefficient of the first detail layer; ω2 is the weight coefficient of the second detail layer.

5. The underwater low-light image enhancement method based on local color correction and multi-scale variational decomposition model according to claim 1, characterized in that: In step 4, the processed base layer and detail layer are combined by linear fusion to generate the optimized V channel as follows: Among them, V E is the optimized V channel; Represents the enhanced base layer; and They are represented as the enhanced first detail layer and second detail layer respectively.

Images