Perception-consistent underwater image enhancement method

Abstract

The invention provides a perception-consistent underwater image enhancement method, and relates to the technical field of image enhancement. The method comprises the following steps: firstly, performing dual-channel color compensation, white balance correction and histogram contrast enhancement on an original image to obtain a color enhanced image; secondly, performing median filtering and gamma correction on a brightness channel to generate a multi-exposure image sequence; extracting a detail layer and a base layer of each exposure image, performing weighted fusion based on contrast, exposure and saturation weights, and realizing detail enhancement in combination with rapid steering filtering; and finally, fusing the color enhanced image and the detail enhanced image according to the contrast ratio, the saturation degree and the structural saliency. According to the method, the image quality is improved through dual-path cooperation and dynamic exposure fusion, the method is suitable for scenes such as underwater monitoring, aquatic product monitoring and underwater inspection, complex training is not needed, and deployment on low-power-consumption equipment is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of image enhancement, and in particular to a method for enhancing underwater images with consistent perception. Background Art

[0002] With the rapid development of smart agriculture and aquaculture technologies, underwater visual sensing has become a key component of the new generation of agricultural consumer electronics, widely used in scenarios such as aquaculture environmental monitoring, hydroponic system testing, and agricultural ecological assessment. However, due to water's wavelength-selective absorption and scattering of light, underwater images often exhibit significant quality degradation, including color deviation, uneven brightness, low contrast, and loss of detail. These degradation issues not only affect effective human perception but also severely interfere with subsequent automated analysis tasks (such as target recognition and behavior analysis). This is particularly prominent on resource-constrained, low-power embedded platforms.

[0003] To address these issues, existing methods often rely on deep learning or complex modeling, which typically requires extensive computing resources and training data, making them difficult to deploy directly on resource-constrained edge computing devices. Therefore, there is an urgent need for an underwater image processing method with high computational efficiency, simple structure, and stable enhancement effects to meet the requirements of low power consumption, real-time performance, and interpretability in agricultural electronic equipment. Summary of the Invention

[0004] In response to the technical problems mentioned in the above background technology, the present invention provides a perceptually consistent underwater image enhancement method. The present invention designs a dual-channel compensation mechanism, combines the advantages of white balance and channel adjustment, and jointly models global color temperature alignment and local color balance to achieve visually natural and balanced results. For detail enhancement, we are inspired by the sensitivity of the human eye to brightness changes and construct a gamma-modulated multi-exposure image sequence based on the brightness channel. By fusing underexposed images (to enhance bright areas) and overexposed images (to restore dark details), this method effectively expands the dynamic brightness range, improves detail expression capabilities, and preserves local structural content, thereby enhancing global clarity and perceptual authenticity.

[0005] The technical means adopted in the present invention are as follows:

[0006] A perceptually consistent underwater image enhancement method comprises the following steps:

[0007] Step S01: converting the original image into the CIELab space, obtaining the A and B channels of the original image in the CIELab space, and performing chromaticity compensation on the A and B channels;

[0008] Step S02: converting the original image in the CIELab space after chromaticity channel compensation to the RGB space, and performing white balance correction and histogram contrast stretching on the original image converted to the RGB space to obtain a color-corrected image;

[0009] Step S03: performing median filtering denoising and histogram stretching brightness processing on the L channel of the color-corrected image, and then performing gamma modulation on the brightness-processed L channel to generate a multi-exposure sequence image; the multi-exposure sequence image includes: an overexposed image, a normal image, and an underexposed image;

[0010] Step S04: A low-frequency base layer is extracted from each multi-exposure sequence image using Gaussian filtering, and a high-frequency detail layer is extracted from each exposure image using residual calculation. The contrast weight, exposure weight, and saturation weight of each exposure image are then calculated. The high-frequency detail layer and the low-frequency base layer are fused based on the contrast weight, exposure weight, and saturation weight. The fused result is then fed into a fast guided filter to obtain a detail-enhanced image.

[0011] Step S05: Perform Laplacian pyramid decomposition on the color-corrected image obtained in step S02 and the detail-enhanced image obtained in step S04 to generate multi-scale levels; fuse the color-corrected image and the detail-enhanced image at each level according to the adaptive weights, and reconstruct the final image through the Laplacian pyramid inverse transform of the fusion result to obtain the final perceptually consistent enhanced image.

[0012] Furthermore, the chromaticity compensation in step S01 includes the following steps:

[0013] Step S011: converting the original image into CIELab space through Opencv to obtain A and B channel data of the input image;

[0014] Step S012: Perform mean difference calculation based on the A and B channel data of the input image:

[0015]

[0016] Among them, m, n represent the height and width of the image, a ij , b ij represents the chromaticity value at the pixel position (i, j), and Δ represents the average color difference between the A and B channels;

[0017] Step S013: Based on the average color difference, adjust the chromaticity values ​​of the A and B channels to achieve color balance:

[0018]

[0019] Among them, α represents the compensation factor for controlling color shift, a and n represent the chromaticity values ​​of A and B channels respectively. Indicates the chromaticity value of the A and B channels after chromaticity balance is completed, and the intermediate result image I after chromaticity balance is achieved bal .

[0020] Furthermore, the white balance and histogram contrast adjustment in step S02 includes the following steps:

[0021] Step S021: Calculate the mean μ of each channel r 、μ g 、μ b :

[0022]

[0023] Among them, m and n represent the width and height μ of the image r 、μ g 、μ b Represents the mean of the three RGB channels respectively;

[0024] Step S022: define the normalization factor k:

[0025]

[0026] Where k represents the normalized scaling factor;

[0027] Step S023: Limit the soft threshold vector s to [0, 0.5]:

[0028] s=clip(0.005*k,,0,,0.5);

[0029] Where s represents the soft threshold vector, which controls the amplitude of dynamic range compression.

[0030] Furthermore, the brightness channel preprocessing in step S03 includes the following steps:

[0031] Step S031: The median filter denoising formula for the color-corrected image is:

[0032] L filtered (x,y)=median({L(x′,y′)|(x′,y′)∈neighborhood(x,y)});

[0033] Among them, neighborhood(x,y) represents the local neighborhood centered at (x,y), median represents the median of all pixel values ​​in the neighborhood, and L filtered (x,y) represents the brightness channel after median filtering;

[0034] Step S032: The histogram stretching formula for the color correction image is:

[0035]

[0036] Among them, L min and L max Indicates the minimum and maximum values ​​in the L channel after filtering.

[0037] Furthermore, the step S04 includes the following steps:

[0038] Step S041: The formula for extracting details from the multi-exposure sequence image is:

[0039]

[0040] in, Represents the low-frequency information after Gaussian filtering, I k represents the kth exposure version of the input image, * represents the convolution operation, G σ represents a Gaussian kernel with standard deviation σ;

[0041] Step S042: The formula for calculating the contrast weight of the multi-exposure sequence images is:

[0042]

[0043] Among them, Y i represents the brightness channel of the i-th exposure image; K x , K y Represents the horizontal and vertical convolution kernels; represents the contrast weight map of the i-th exposure image;

[0044] Step S043: The formula for calculating the exposure weight of the multi-exposure sequence image is:

[0045] L i =log(max(Y i ,ε));

[0046]

[0047] Among them, Y i represents the brightness channel of the i-th image; ε represents a very small constant to avoid the logarithm from taking 0; L i represents the brightness channel after logarithmic transformation; H and W represent the height and width of the image; μ L Represents the brightness channel L i The global mean of ; σ represents the standard deviation of the Gaussian function; represents the exposure weight map;

[0048] Step S044: The formula for calculating the saturation weight of the exposure image is:

[0049]

[0050] Among them, U i and V i Represents the chrominance component of pixel (x, y) in the i-th exposure image.

[0051] Furthermore, in the contrast weight calculation in step S042 , Kx=[−1, 1], Ky=[−1, 1]T.

[0052] Furthermore, the minimum constant ε for calculating the exposure weight in step S043 is 10-8.

[0053] Furthermore, in the extraction calculation of the exposure weight in step S043, σ=0.2.

[0054] The present invention achieves perceptual consistency in color structure through dual-channel compensation and multi-exposure fusion detail enhancement. Through chromaticity compensation, the present invention effectively eliminates the common blue-green and local color casts in underwater images. Compared with existing technologies, the RGB channel distribution is nearly balanced, avoiding problems such as oversaturation and color distortion. Furthermore, the present invention generates multi-exposure sequences through gamma modulation, emphasizing the extraction of details and preserving edge structures. Key point detection experiments show that this method's detail recovery capability is superior to traditional methods. Finally, the present invention is based on traditional image processing workflows, does not require pre-trained models or big data support, has low overhead, low module complexity, and supports low-efficiency sustainable computing.

[0055] Based on the above reasons, the present invention can be promoted and applied in the fields of digital image processing and the like. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0057] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0058] Figure 2 This is a comparison chart of the shipwreck image enhancement effects of the present invention and other algorithms. Figure 2-1 Indicates the original image. Figure 2-2 Result graph showing DCP; Figure 2-3 Result graph showing UDCP; Figure 2-4Graph showing the results of IBLA; Figure 2 -5 indicates the result graph of ULAP; Figure 2 -6 is a graph showing the results of the method of the present invention.

[0059] Figure 3 This is a comparison chart of the fish school image enhancement effects of the present invention and other algorithms. Figure 3-1 Indicates the original image. Figure 3-2 Result graph showing DCP; Figure 3-3 Result graph showing UDCP; Figure 3-4 Graph showing the results of IBLA; Figure 3 -5 indicates the result graph of ULAP; Figure 3 -6 is a graph showing the results of the method of the present invention.

[0060] Figure 4 This is a comparison chart of the effects of the present invention and other algorithms on archaeological image enhancement. Figure 4-1 Indicates the original image. Figure 4-2 Result graph showing DCP; Figure 4-3 Result graph showing UDCP; Figure 4-4 Graph showing the results of IBLA; Figure 4 -5 indicates the result graph of ULAP; Figure 4 -6 is a graph showing the results of the method of the present invention. DETAILED DESCRIPTION

[0061] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions 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 embodiments described 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 making creative efforts should fall within the scope of protection of the present invention.

[0062] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0063] To verify the effectiveness of the underwater image enhancement method, we selected images from different scenes as test datasets and compared the experimental results with those of DCP (Single image haze removal using dark channel prior), UDCP (Transmission estimation in underwater single images), IBLA (Underwater image restoration based on image blurriness and light absorption), ULAP (A Rapid Scene Depth Estimation Model Based on Underwater Light Attenuation Prior for Underwater Image Restoration), and the present invention through both subjective and objective comparative analysis. The specific steps and principles are as follows:

[0064] like Figure 1 As shown, the present invention provides a method for enhancing underwater images with consistent perception, comprising the following steps:

[0065] Step S01: converting the original image into the CIELab space, obtaining the A and B channels of the original image in the CIELab space, and performing chromaticity compensation on the A and B channels; the chromaticity compensation in step S01 includes the following steps:

[0066] Step S011: converting the original image into CIELab space through Opencv to obtain A and B channel data of the input image;

[0067] Step S012: Perform mean difference calculation based on the A and B channel data of the input image:

[0068]

[0069] Among them, m, n represent the height and width of the image, a ij , b ij represents the chromaticity value at the pixel position (i, j), and Δ represents the average color difference between the A and B channels;

[0070] Step S013: Based on the average color difference, adjust the chromaticity values ​​of the A and B channels to achieve color balance:

[0071]

[0072] Among them, α represents the compensation factor for controlling color shift, a and b represent the chromaticity values ​​of A and B channels respectively. Indicates the chromaticity value of the A and B channels after chromaticity balance is completed, and the intermediate result image I after chromaticity balance is achieved bal .

[0073] Step S02: Converting the original image in the CIELab space after chromaticity channel compensation to the RGB space, and performing white balance correction and histogram contrast stretching on the original image converted to the RGB space to obtain a color-corrected image; the white balance and histogram contrast adjustment in step S02 includes the following steps:

[0074] Step S021: Calculate the mean μ of each channel r 、μ g 、μ b :

[0075]

[0076] Among them, m and n represent the width and height μ of the image r 、μ g 、μ b Represents the mean of the three RGB channels respectively;

[0077] Step S022: define the normalization factor k:

[0078]

[0079] Where k represents the normalized scaling factor;

[0080] Step S023: Limit the soft threshold vector s to [0, 0.5]:

[0081] s=clip(0.005*k,,0,,0.5);

[0082] Where s represents the soft threshold vector, which controls the amplitude of dynamic range compression.

[0083] Step S03: Perform median filtering denoising and histogram stretching brightness processing on the L channel of the color-corrected image, and then perform gamma modulation on the brightness-processed L channel to generate a multi-exposure sequence image; the multi-exposure sequence image includes: an overexposed image, a normal image, and an underexposed image; the brightness channel preprocessing of step S03 includes the following steps:

[0084] Step S031: The median filter denoising formula for the color-corrected image is:

[0085] L filtered (x,y)=median({L(x′,y′)|(x′,y′)∈neighborhood(x,y)});

[0086] Among them, neighborhood(x,y) represents the local neighborhood centered at (x,y), median represents the median of all pixel values ​​in the neighborhood, and L filtered (x,y) represents the brightness channel after median filtering;

[0087] Step S032: The histogram stretching formula for the color correction image is:

[0088]

[0089] Among them, L min and L max Indicates the minimum and maximum values ​​in the L channel after filtering.

[0090] Step S04: extracting a low-frequency base layer from each multi-exposure sequence image through Gaussian filtering, extracting a high-frequency detail layer from each exposure image through residual calculation, and then calculating the contrast weight, exposure weight, and saturation weight of each exposure image. The high-frequency detail layer and the low-frequency base layer are fused according to the contrast weight, exposure weight, and saturation weight, and the fused result is input into a fast guided filter to obtain a detail-enhanced image. Step S04 includes the following steps:

[0091] Step S041: The formula for extracting details from the multi-exposure sequence image is:

[0092]

[0093] in, Represents the low-frequency information after Gaussian filtering, I k represents the kth exposure version of the input image, * represents the convolution operation, G σ represents a Gaussian kernel with standard deviation σ;

[0094] Step S042: The formula for calculating the contrast weight of the multi-exposure sequence images is:

[0095] G x =|Y i *K x |;

[0096] G y =|Y i *K y |;

[0097]

[0098] Among them, Y i represents the brightness channel of the i-th exposure image; K x , K y Represents the horizontal and vertical convolution kernels; Represents the contrast weight map of the i-th exposure image; K in the contrast weight calculation x =[-1,1],K y =[-1,1] T .

[0099] Step S043: The formula for calculating the exposure weight of the multi-exposure sequence image is:

[0100] L i =log(max(Y i ,ε));

[0101]

[0102] Among them, Y i Represents the brightness channel of the i-th image; ε represents a very small constant, ε = 10 -8 , to avoid the logarithm from taking 0; L i represents the brightness channel after logarithmic transformation; H and W represent the height and width of the image; μ L Represents the brightness channel L i The global mean of ; σ represents the standard deviation of the Gaussian function; represents the exposure weight map;

[0103] Step S044: The formula for calculating the saturation weight of the exposure image is:

[0104]

[0105] Among them, U i and V i Represents the chrominance component of pixel (x, y) in the i-th exposure image.

[0106] Step S05: Perform Laplacian pyramid decomposition on the color-corrected image obtained in step S02 and the detail-enhanced image obtained in step S04 to generate multi-scale levels; fuse the color-corrected image and the detail-enhanced image at each level according to the adaptive weights, and reconstruct the final image through the Laplacian pyramid inverse transform of the fusion result to obtain the final perceptually consistent enhanced image.

[0107] Example 1

[0108] like Figure 2Figure 2 shows a comparison of the enhancement effects of the proposed method and existing typical underwater image enhancement algorithms on the same original image. To validate the effectiveness of the proposed method, comparative experiments were conducted using representative algorithms, including IBLA, GDCP, WaterNet, and FunIE-GAN, encompassing both traditional physical model methods and deep learning approaches. The image comparisons reveal that the IBLA and GDCP algorithms still exhibit an overall green cast and low contrast in the processed images, failing to effectively restore true colors. While the WaterNet algorithm partially mitigates the color cast, the image lacks contrast, blurs edge structures, and preserves limited detail. While the FunIE-GAN algorithm exhibits some advantages in contrast enhancement, it still suffers from significant color shift and unnatural image tones. In comparison, the proposed method performs better in color restoration, brightness balance, and structural clarity. The processed image exhibits natural, saturated colors, effectively enhanced dark areas, and sharper edge contours. This significantly improves the color cast, blurring, and low contrast issues of underwater images, resulting in higher perceptual consistency and structural restoration capabilities, validating the effectiveness and practicality of the proposed enhancement strategy.

[0109] like Figure 3 As shown in the figure, the enhancement effect of the method of the present invention on underwater biological images is compared with that of various existing algorithms. From the image effect analysis, it can be seen that the IBLA and WaterNet algorithms have obvious color imbalance problems, the overall image is purple, and the true color is not effectively restored; the image brightness after processing by the GDCP algorithm is obviously too high, there is overexposure, resulting in loss of image details; the FunIE-GAN algorithm improves the image contrast to a certain extent, but the overall image is dark, and the visibility of the underwater target area is poor. In contrast, the method of the present invention shows higher color restoration and brightness balance in the fish area. The enhanced image has natural colors, clear edges, and significantly improved contrast, effectively preserving the details and structural information of the fish school. The above results show that the method of the present invention can significantly improve the color cast, brightness imbalance and detail blurring problems of underwater images, has excellent visual perception consistency and detail enhancement capabilities, and is suitable for practical application scenarios such as underwater biological recognition and target detection.

[0110] like Figure 4 The following figure compares the enhancement effects of the proposed method and a typical existing algorithm on seabed target detail areas. The experimental scene focuses on the seabed structure area with rich details in the image to evaluate the performance of each algorithm in contrast enhancement and detail recovery.

[0111] From the image effect analysis, it can be seen that the IBLA algorithm enhances the image contrast to a certain extent, but the visual improvement is not significant, and the detail areas are still blurred; although the GDCP algorithm improves the contrast significantly, the overall image is blue and there is an obvious color cast problem; the WaterNet algorithm performs relatively balanced in color correction and contrast adjustment, but the detail clarity is insufficient and the image is still smooth locally; the FunIE-GAN algorithm has serious color cast and low contrast, which makes it impossible to accurately restore structural details. In contrast, the method of the present invention shows higher contrast and detail restoration capabilities in the enhanced image. The image color is natural, the structural edges are clear, and the detail layering is strong, which significantly improves the shortcomings of the existing algorithm in color cast and detail loss. The experimental results verify the dual improvement effect of the method on visual quality and structural perception ability in underwater image enhancement tasks, and is suitable for practical application scenarios such as underwater target recognition and seabed resource surveying.

[0112] To avoid analytical bias caused by relying solely on subjective visual evaluation, this example introduces three typical objective evaluation metrics: PSNR (peak signal-to-noise ratio), SSIM (structural similarity), and UCIQE (underwater image quality evaluation index). This comparative evaluation evaluates the performance of the proposed method with existing algorithms such as IBLA, GDCP, WaterNet, and FunIE-GAN in underwater image enhancement tasks. As shown in Table 1, the proposed method achieves the highest values ​​for all three metrics, demonstrating superior image enhancement quality.

[0113] Specifically, this invention uses a dual-channel color compensation mechanism to effectively correct global and local color casts in underwater images, significantly improving color reproduction. Simultaneously, a multi-exposure-guided detail enhancement strategy strengthens the structural contrast between bright and dark areas, resulting in sharper edges and more distinct layers. The combination of these two approaches ensures a natural color balance in images while significantly enhancing detail preservation.

[0114] In summary, the method of the present invention not only improves the perceived quality of images, but also takes into account structural consistency and contrast enhancement. Compared with existing methods, it has higher PSNR, SSIM and UCIQE values, which verifies its superior performance and wide applicability in underwater image enhancement and restoration tasks.

[0115] Table 1 Comparison of evaluation values ​​of the processing results of the algorithm of the present invention and other algorithms

[0116]

[0117] The serial numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts not described in detail in a particular embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented by other means.

[0118] 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 above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above 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. A perceptually consistent underwater image enhancement method, characterized in that: The following steps are involved: Step S01: converting the original image into the CIELab space, obtaining the A and B channels of the original image in the CIELab space, and performing chromaticity compensation on the A and B channels; Step S02: converting the original image in the CIELab space after chromaticity channel compensation to the RGB space, and performing white balance correction and histogram contrast stretching on the original image converted to the RGB space to obtain a color-corrected image; Step S03: performing median filtering denoising and histogram stretching brightness processing on the L channel of the color-corrected image, and then performing Gamma modulation on the brightness-processed L channel to generate a multi-exposure sequence image; The multi-exposure sequence images include: an overexposed image, a normal image, and an underexposed image; Step S04: A low-frequency base layer is extracted from each multi-exposure sequence image using Gaussian filtering, and a high-frequency detail layer is extracted from each exposure image using residual calculation. The contrast weight, exposure weight, and saturation weight of each exposure image are then calculated. The high-frequency detail layer and the low-frequency base layer are fused based on the contrast weight, exposure weight, and saturation weight. The fused result is then fed into a fast guided filter to obtain a detail-enhanced image. Step S05: Perform Laplacian pyramid decomposition on the color-corrected image obtained in step S02 and the detail-enhanced image obtained in step S04 to generate multi-scale levels; fuse the color-corrected image and the detail-enhanced image at each level according to the adaptive weights, and reconstruct the final image through the Laplacian pyramid inverse transform of the fusion result to obtain the final perceptually consistent enhanced image.

2. The perceptually consistent underwater image enhancement method according to claim 1, further comprising: The chromaticity compensation in step S01 includes the following steps: Step S011: converting the original image into CIELab space through Opencv to obtain A and B channel data of the input image; Step S012: Perform mean difference calculation based on the A and B channel data of the input image: Among them, m, n represent the height and width of the image, a ij , b ij represents the chromaticity value at the pixel position (i, j), and Δ represents the average color difference between the A and B channels; Step S013: Based on the average color difference, adjust the chromaticity values ​​of the A and B channels to achieve color balance: Among them, α represents the compensation factor for controlling color shift, a and b represent the chromaticity values ​​of A and B channels respectively. Indicates the chromaticity value of the A and B channels after chromaticity balance is completed, and the intermediate result image I after chromaticity balance is achieved bal .

3. The perceptually consistent underwater image enhancement method according to claim 1, further comprising: The white balance and histogram contrast adjustment in step S02 includes the following steps: Step S021: Calculate the mean μ of each channel r 、μ g 、μ b : Among them, m and n represent the width and height μ of the image r 、μ g 、μ b Represents the mean of the three RGB channels respectively; Step S022: define the normalization factor k: Where k represents the normalized scaling factor; Step S023: Limit the soft threshold vector s to [0, 0.5]: s=clip(0.005*k,,0,,0.5); Where s represents the soft threshold vector, which controls the amplitude of dynamic range compression.

4. The perceptually consistent underwater image enhancement method according to claim 1, further comprising: The brightness channel preprocessing of step S03 includes the following steps: Step S031: The median filter denoising formula for the color-corrected image is: L filtered (x,y)=median({L(x′,y′)|(x′,y′)∈neighborhood(x,y)}); Among them, neighborhood(x,y) represents the local neighborhood centered at (x,y), median represents the median of all pixel values ​​in the neighborhood, and L filtered (x,y) represents the brightness channel after median filtering; Step S032: The histogram stretching formula for the color correction image is: Among them, L min and L max Indicates the minimum and maximum values ​​in the L channel after filtering.

5. The perceptually consistent underwater image enhancement method according to claim 1, further comprising: The step S04 includes the following steps: Step S041: The formula for extracting details from the multi-exposure sequence image is: in, Represents the low-frequency information after Gaussian filtering, I k represents the kth exposure version of the input image, * represents the convolution operation, G σ represents a Gaussian kernel with standard deviation σ; Step S042: The formula for calculating the contrast weight of the multi-exposure sequence images is: G x =|Y i *K x |; G y =|Y i *K y |; Among them, Y i represents the brightness channel of the i-th exposure image; K x , K y Represents the horizontal and vertical convolution kernels; represents the contrast weight map of the i-th exposure image; Step S043: The formula for calculating the exposure weight of the multi-exposure sequence image is: L i =log(max(Y i ,e)); Among them, Y i represents the brightness channel of the i-th image; ε represents a very small constant to avoid the logarithm from taking 0; L i represents the brightness channel after logarithmic transformation; H and W represent the height and width of the image; μ L Represents the brightness channel L i The global mean of ; σ represents the standard deviation of the Gaussian function; represents the exposure weight map; Step S044: The formula for calculating the saturation weight of the exposure image is: Among them, U i and V i Represents the chrominance component of pixel (x, y) in the i-th exposure image.

6. The perceptually consistent underwater image enhancement method according to claim 5, further comprising: The K in the contrast weight calculation in step S042 x =[-1,1],K y =[-1,1] T .

7. The perceptually consistent underwater image enhancement method according to claim 5, further comprising: The minimum constant ε in the exposure weight calculation in step S043 is 10 -8 .

8. The method for enhancing underwater images with consistent perception according to claim 5, wherein: In the step S043 , the exposure weight σ is extracted and calculated as 0.2.

Images