Self-adaptive underwater image enhancement method based on multi-domain fusion and color compensation

Abstract

The invention discloses a self-adaptive underwater image enhancement method based on multi-domain fusion and color compensation, and belongs to the technical field of image processing, and the method comprises the following steps: S1, obtaining an underwater image, and carrying out the data preprocessing of the underwater image; s2, carrying out adaptive underwater defogging compensation processing on the preprocessed underwater image; and S3, performing color balance processing on the underwater image after the adaptive underwater defogging compensation processing to complete underwater image enhancement. According to the method, the defects of insufficient cross-environment generalization ability, inaccurate color recovery, large calculation burden and the like in the prior art are effectively solved, and the method has better underwater environment adaptability and real-time processing potential in combination with self-adaption, a physical principle-based module, selective filtering and a color optimization strategy; and a more accurate and efficient technical scheme is provided for fine perception of the underwater environment.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to an adaptive underwater image enhancement method based on multi-domain fusion and color compensation. Background Technology

[0002] With the rapid development of applications in marine resource exploration, seabed archaeology, ecological monitoring, and underwater robotics, underwater image acquisition and analysis are playing an increasingly important role in various marine missions. Since approximately 70% to 80% of information acquisition relies on visual input, clear and reliable images are crucial for human and machine environmental perception, target recognition, and path planning. However, due to the complex underwater imaging conditions, acquiring high-quality underwater images still faces numerous technical challenges.

[0003] Underwater optical images typically suffer from several degradation problems: First, due to the significant wavelength dependence of light absorption in water, red light attenuates rapidly over short distances, resulting in images generally exhibiting a blue-green hue. Second, forward and backscattering effects caused by suspended particles in the water significantly reduce image contrast and lead to blurred details. Furthermore, uneven distribution of ambient light sources, variations in seawater turbidity, and differences in water depth further exacerbate the complexity of image degradation.

[0004] To address the aforementioned issues, traditional underwater image enhancement methods mainly include histogram equalization, Retinex theory, dark channel prior (DCP), and filtering-based enhancement techniques. While these methods can improve image contrast or enhance brightness uniformity in specific scenarios, they perform poorly in highly turbid or extremely low-light environments, often resulting in loss of detail, color distortion, or over-enhancement. Furthermore, some physically based algorithms, such as the Jaffe–McGlamery model and the Sea-thru method, while possessing strong interpretability, rely on high-precision depth information or specific environmental assumptions, making them difficult to generalize to widespread applications.

[0005] In recent years, with the development of deep learning technology, more and more studies have attempted to achieve automatic underwater image enhancement through convolutional neural networks or generative adversarial networks. These methods have achieved excellent performance on standard datasets, but still have the following shortcomings: First, they are highly dependent on large-scale labeled training sets, making it difficult to obtain complete and realistic paired data in actual marine environments; second, the network structure often lacks interpretability, making it difficult to guarantee the physical rationality of the enhancement process; and third, most models have high computational complexity and high resource consumption, making them unsuitable for the real-time processing needs of embedded underwater platforms.

[0006] In summary, existing technologies are insufficient to simultaneously meet the comprehensive requirements of adaptability to multiple types of degradation, physical interpretability, low resource requirements, and color fidelity. There is an urgent need for an underwater image enhancement method that combines physical model guidance and perception optimization capabilities with high efficiency and deployability. Summary of the Invention

[0007] To address the above problems, this invention proposes an adaptive underwater image enhancement method based on multi-domain fusion and color compensation.

[0008] The technical solution of this invention is: an adaptive underwater image enhancement method based on multi-domain fusion and color compensation, comprising the following steps:

[0009] S1. Acquire underwater images and perform data preprocessing on the underwater images;

[0010] S2. Perform adaptive underwater defogging compensation processing on the preprocessed underwater image;

[0011] S3. Perform color balancing on the underwater image after adaptive underwater defogging compensation to complete underwater image enhancement.

[0012] Furthermore, S1 includes the following sub-steps:

[0013] S11. Based on the enhanced image, generate a preprocessed underwater image using a wavelet-guided adaptive filtering method.

[0014] S12. Based on the enhanced image, generate a preprocessed image using a wavelet-guided adaptive filtering strategy.

[0015] Furthermore, in S11, the enhanced image I hmim The formula for calculating (x,y) is:

[0016]

[0017] L(x,y)=α(x,y)·L global (x,y)+(1-α(x,y))·L local (x,y);

[0018]

[0019]

[0020] Where x represents the x-coordinate, y represents the y-coordinate, and L... global (x,y) represents the global brightness estimate, L local (x,y) represents the local brightness estimate. I represents the result after applying global gamma correction to the input image. in(x,y) represents the intensity of the original gray value of the input image at coordinates (x,y), and S represents the number of layers in the multi-scale Gaussian filter. The scale is represented by σ. s Gaussian kernel, ω s Let L(x,y) represent the corresponding weight coefficients, L(x,y) represent the final illumination map, α(x,y) represent the adaptive weights at each location in the image, and ε represent a very small positive number.

[0021] Furthermore, S12 includes the following sub-steps:

[0022] S121. Use discrete wavelet transform to decompose the enhanced image into several frequency sub-bands;

[0023] S122. In each frequency sub-band, determine the noise level based on the local variance;

[0024] S123. Perform bilateral filtering on low-noise-level regions and high-pass filtering on high-noise-level regions; where high-noise-level regions are local areas with a signal-to-noise ratio (SNR) below 20dB, and low-noise-level regions are regions with a SNR above or equal to 20dB; SNR is calculated by the ratio of local signal power to noise power.

[0025] S124. Combine the filtered frequency subbands to generate a preprocessed underwater image.

[0026] Furthermore, in S122, the noise level N est The formula for calculating (x,y) is:

[0027]

[0028] Where Ω represents the neighborhood window, μ Ω Let (x, y) represent the average value within the window, (x, y) represent the pixel coordinates of the current noise level to be estimated, and i represent the loop variable when traversing all pixels in the neighborhood window. hmin (i) represents the gray value of the i-th pixel within the neighborhood window.

[0029] Furthermore, S2 includes the following sub-steps:

[0030] S21. Calculate the spectral attenuation ratio based on the preprocessed underwater image;

[0031] S22. Calculate the water body type corresponding to the preprocessed underwater image based on the spectral attenuation ratio;

[0032] S23. Based on the water body type, use the optimization objective function to perform multi-priority fusion compensation to complete the adaptive underwater fogging compensation process.

[0033] Furthermore, in S21, the spectral attenuation ratio ηc The calculation formula is:

[0034]

[0035] Where, μ c μ represents the average brightness of a channel in an image. R μ represents the average brightness of channel R in the image. G μ represents the average brightness of channel G in the image. B This represents the average brightness of channel B in the image;

[0036] In S22, the formula for calculating the water body type (WTI) corresponding to the preprocessed underwater image is as follows:

[0037]

[0038] in, This represents the reference attenuation ratio for the ω-th type of water body;

[0039] In S23, the underwater image I after adaptive underwater defogging compensation processing audc The formula for calculating (x) is:

[0040]

[0041] Where x represents the horizontal coordinate, J represents the objective function or image to be optimized, J(x) represents the function value or pixel value of the objective function or image at a certain location, and J... j d(x) represents the recovery result obtained from the j-th prior, d(x) represents the image depth, and λ represents the depth of the image. j (d(x)) represents the weighting coefficient determined by the image depth and water body type. β represents the total variation regularization term, and β represents the regularization weight parameter.

[0042] Furthermore, S3 includes the following sub-steps:

[0043] S31. Construct a reference color distribution;

[0044] S32. Compare the underwater image after adaptive underwater defogging compensation with the reference color distribution, and calculate the CIEDE2000 color difference of the image at the pixel level.

[0045] S33. Based on the CIEDE2000 color difference at the pixel level, perform color correction to complete underwater image enhancement.

[0046] Furthermore, in S33, the formula for calculating color correction is:

[0047] I pdc (x)=I audc (x)+β·(ΔEtarget (x)-ΔE current (x));

[0048] Where x represents the x-coordinate, I pdc (x) represents the final optimized output value, I audc (x) represents the underwater image after adaptive underwater defogging compensation processing, β represents the correction step size, and ΔE target (x) represents the color difference threshold that is difficult for the human eye to distinguish, ΔE current (x) represents the perceptual error between the current pixel and the reference.

[0049] The beneficial effects of this invention are: This invention effectively solves the shortcomings of existing technologies such as insufficient cross-environment generalization ability, inaccurate color restoration and high computational burden. By combining adaptive, physical principle-based modules with selective filtering and color optimization strategies, it has better underwater environment adaptability and real-time processing potential, providing a more accurate and efficient technical solution for fine underwater environment perception. Attached Figure Description

[0050] Figure 1 The flowchart shows an adaptive underwater image enhancement method based on multi-domain fusion and color compensation.

[0051] Figure 2 Comparison of enhancement results for several representative images in the UIEBD test set;

[0052] Figure 3 This is a typical result graph on the Challenge60 dataset. Detailed Implementation

[0053] The embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0054] like Figure 1 As shown, this invention provides an adaptive underwater image enhancement method based on multi-domain fusion and color compensation, comprising the following steps:

[0055] S1. Acquire underwater images and perform data preprocessing on the underwater images;

[0056] S2. Perform adaptive underwater defogging compensation processing on the preprocessed underwater image;

[0057] S3. Perform color balancing on the underwater image after adaptive underwater defogging compensation to complete underwater image enhancement.

[0058] To mitigate optical image degradation issues encountered in underwater imaging, such as wavelength-dependent attenuation, scattering-induced fog, and noise amplification, this invention proposes an image enhancement framework based on physical principles while also considering visual perception. The framework consists of three stages, each addressing a specific type of degradation in underwater imaging. The order of the stages is designed to reflect their functional interdependence and the inherent characteristics of the degradation phenomena to be corrected.

[0059] In this embodiment of the invention, S1 includes the following sub-steps:

[0060] S11. Construct a hybrid multi-scale illumination model and use it to correct the global and local illumination of underwater images to generate enhanced images;

[0061] S12. Based on the enhanced image, generate a preprocessed underwater image using a wavelet-guided adaptive filtering method.

[0062] This phase primarily addresses two main causes of reduced visibility in underwater images: non-uniform illumination and high-frequency noise. First, a hybrid multi-scale illumination model is employed to correct spatial brightness non-uniformity; this model combines Gamma correction and multi-scale Retinex principles. Subsequently, a wavelet-guided adaptive filtering strategy is used to suppress noise in both the spatial and frequency domains while preserving the detailed texture and structural information required for subsequent enhancement stages.

[0063] In this embodiment of the invention, in S11, firstly, a hybrid multi-scale illumination model is used to correct for global and local illumination variations. Global illumination is adjusted using gamma correction to compensate for the human eye's sensitivity to brightness; while local illumination is optimized using multi-scale Gaussian filtering to adapt to spatially varying lighting conditions. The final illumination map can be represented as a weighted fusion of local and global illumination. After brightness normalization, the enhanced image I... hmim The formula for calculating (x,y) is:

[0064]

[0065] L(x,y)=α(x,y)·L global (x,y)+(1-α(x,y))·L local (x,y);

[0066]

[0067]

[0068] Where x represents the x-coordinate, y represents the y-coordinate, and L... global (x,y) represents the global brightness estimate, L local(x,y) represents the local brightness estimate. I represents the result after applying global gamma correction to the input image. in (x,y) represents the intensity of the original gray value of the input image at coordinates (x,y), and S represents the number of layers in the multi-scale Gaussian filter. The scale is represented by σ. s Gaussian kernel, ω s Let L(x,y) represent the corresponding weight coefficients, L(x,y) represent the final illumination map, α(x,y) represent the adaptive weights at each location in the image, and ε represent a small positive number. The purpose of ε is to avoid division by zero errors when L(x,y) is close to 0 in some regions.

[0069] In this embodiment of the invention, S12 includes the following sub-steps:

[0070] S121. Use discrete wavelet transform to decompose the enhanced image into several frequency sub-bands;

[0071] S122. In each frequency sub-band, determine the noise level based on the local variance;

[0072] S123. Perform bilateral filtering on low-noise-level regions and high-pass filtering on high-noise-level regions; where high-noise-level regions are local areas with a signal-to-noise ratio (SNR) below 20dB, and low-noise-level regions are regions with a SNR above or equal to 20dB; SNR is calculated by the ratio of local signal power to noise power.

[0073] S124. Combine the filtered frequency subbands to generate a preprocessed underwater image.

[0074] In this embodiment of the invention, in S122, this operation makes the overall brightness of the image more uniform, and the details are preserved, facilitating subsequent processing. To reduce spatial and frequency domain noise caused by environmental interference and imaging devices, a wavelet-guided adaptive filtering strategy is adopted. The specific method is as follows: First, the input image is decomposed into multiple frequency sub-bands using discrete wavelet transform, and low-frequency background and high-frequency details are extracted respectively. Second, in each high-frequency sub-band, the noise level of the region is estimated by calculating the local variance. Then, bilateral filtering is used for low-noise regions to maintain edge sharpness, while high-noise regions are enhanced with high-pass filters. Finally, the filtered frequency sub-bands are recombined into a denoised image, thereby preserving edge and texture information. Noise level N est The formula for calculating (x,y) is:

[0075]

[0076] Where Ω represents the neighborhood window, μ ΩLet (x, y) represent the average value within the window, (x, y) represent the pixel coordinates of the current noise level to be estimated, and i represent the loop variable when traversing all pixels in the neighborhood window. hmin (i) represents the gray value of the i-th pixel within the neighborhood window.

[0077] In this embodiment of the invention, S2 includes the following sub-steps:

[0078] S21. Calculate the spectral attenuation ratio based on the preprocessed underwater image;

[0079] S22. Calculate the water body type corresponding to the preprocessed underwater image based on the spectral attenuation ratio;

[0080] S23. Based on the water body type, use the optimization objective function to perform multi-priority fusion compensation to complete the adaptive underwater fogging compensation process.

[0081] After achieving uniform illumination and denoising of the image, a physics-based dehazing process is applied to restore the image's natural color fidelity. An adaptive underwater dehazing compensation algorithm is proposed, which can directly estimate spectral attenuation characteristics from the image and adaptively compensate according to water body type classification. This stage is placed after denoising to avoid amplifying noise and ensure accurate modeling of the underwater light propagation process.

[0082] In this embodiment of the invention, in S21, according to Beer-Lambert's law, light of different wavelengths attenuates at different rates in water, resulting in red light attenuating the fastest in the image, while the blue and green channels dominate. Therefore, AUDC first estimates the spectral attenuation ratio by normalizing the mean intensity of the three RGB channels of the image. The spectral attenuation ratio η c The calculation formula is:

[0083]

[0084] Where, μ c μ represents the average brightness of a channel in an image. R μ represents the average brightness of channel R in the image. G μ represents the average brightness of channel G in the image. B This represents the average brightness of channel B in the image;

[0085] In S22, to improve the adaptability of the compensation, the system compares the estimated attenuation vector with multiple reference templates in the Jerlov water classification model to identify the water body type corresponding to the current image. The formula for calculating the water body type (WTI) corresponding to the preprocessed underwater image is as follows:

[0086]

[0087] in, This represents the reference attenuation ratio for the ω-th type of water body;

[0088] In S23, based on the objectives of dehazing and color restoration, three classic image priors are introduced: dark channel prior, red channel compensation, and multi-scale underwater DCP, respectively used for different depth regions and degradation levels. To avoid the limitations of a single prior, the following optimization objective function is proposed to achieve multi-prior fusion. Underwater image I after adaptive underwater dehazing compensation processing. audc The formula for calculating (x) is:

[0089]

[0090] Where x represents the horizontal coordinate, J represents the objective function or image to be optimized, J(x) represents the function value or pixel value of the objective function or image at a certain location, and J... j d(x) represents the recovery result obtained from the j-th prior, d(x) represents the image depth, and λ represents the depth of the image. j (d(x)) represents the weighting coefficient determined by the image depth and water body type. β represents the total variation regularization term, and β represents the regularization weight parameter.

[0091] In this embodiment of the invention, S3 includes the following sub-steps:

[0092] S31. Construct a reference color distribution;

[0093] S32. Compare the underwater image after adaptive underwater defogging compensation with the reference color distribution, and calculate the CIEDE2000 color difference of the image at the pixel level.

[0094] S33. Based on the CIEDE2000 color difference at the pixel level, perform color correction to complete underwater image enhancement.

[0095] While adaptive underwater fog compensation algorithms can enhance image realism based on physical properties, some perceptual inconsistencies may still exist. A perception-driven color balance algorithm minimizes color deviation in the LAB color space using the CIEDE2000 color difference metric, thereby optimizing the final output to better align with human visual preferences. This ensures that the image possesses both physical accuracy and a good visual appeal.

[0096] This structured processing sequence allows each stage to build upon the previous one, resulting in a synergistic and enhanced overall effect. By systematically correcting illumination, removing noise, restoring color, and optimizing perceived quality, the overall framework balances image interpretability and processing performance in a variety of underwater environments.

[0097] In this embodiment of the invention, the calculation formula for color correction in S33 is as follows:

[0098] I pdc (x)=I audc (x)+β·(ΔE target (x)-ΔE current (x));

[0099] Where x represents the x-coordinate, I pdc (x) represents the final optimized output value, I audc (x) represents the underwater image after adaptive underwater defogging compensation processing, β represents the correction step size, and ΔE target (x) represents the color difference threshold that is difficult for the human eye to distinguish, ΔE current (x) represents the perceptual error between the current pixel and the reference.

[0100] In this embodiment of the invention, to verify the effectiveness and adaptability of the invention, extensive experiments were conducted based on multiple publicly available underwater image datasets, including the UIEBD dataset, the U45 dataset, and the Challengering60 subset, which contains severe degradation scenes. All images were uniformly scaled to standard resolution and underwent uniform preprocessing to eliminate the influence of differences in external imaging equipment.

[0101] This embodiment compares the method proposed in this invention with six existing representative algorithms, including EUIVF (fusion algorithm), OCM (color mapping), UDCP (dark channel prior), TSA (transfer learning), FGAN (generative adversarial network), and DLIFM (deep guided fusion method), covering traditional models, data-driven methods, and hybrid strategies.

[0102] Figure 2 The presentation shows a comparison of enhancement results for several representative images in the UIEBD test set. The results indicate that traditional methods (such as UDCP and OCM) exhibit significant color deviation, blurred edges, and loss of some details in highly scattering water. While deep learning methods (such as FGAN and DLIFM) perform well under specific conditions, they suffer from color oversaturation or artifact enhancement issues under extreme conditions. The method proposed in this invention achieves more natural color restoration, higher contrast, and better detail preservation in multiple test images, with particularly significant advantages in deep water or highly turbid areas.

[0103] Figure 3 The presentation showcases typical results on the Challenge60 dataset, a subset containing numerous severely degraded images (left), such as those in low-light, strong scattering, and large suspended particle environments, realistically simulating complex underwater scenes. The comparison image shows that the present invention (right) enhances image sharpness, effectively avoiding issues like "bluishness," "overexposure," and "false contours," resulting in an enhanced image that more closely resembles natural visual perception.

[0104] To further verify the performance of this invention, three industry-recognized underwater image quality evaluation indicators were used:

[0105] UCIQE: Measures color quality, contrast, and saturation;

[0106] UIQM: A comprehensive evaluation of image color, sharpness, and texture;

[0107] UISM: Evaluates the degree to which image structure and edge information are preserved.

[0108] Table 1 shows the comparison results of different methods on the UIEBD dataset under three metrics.

[0109] Table 1

[0110] method UCIQE UIQM UISM EUIVF 0.6151 3.0803 5.3508 OCM 0.6765 3.8784 5.1003 UDCP 0.5852 3.3724 4.6308 TSA 0.5587 3.9101 4.9002 FGAN 0.5379 3.4710 5.0426 DLIFM 0.6087 3.1661 5.3308 This invention 0.6853 3.8408 5.9417

[0111] The results show that the present invention achieves leading performance on both the standard set and the difficult subset of UIEBD, with an improvement of about 11.4% in UCIQE and about 9.5% in UISM, achieving a unity of color realism and perceptual balance while maintaining edge structure.

[0112] On the U45 dataset, the present invention also obtained the following values. Table 2 shows the comparison results of different methods on the U45 dataset under three metrics.

[0113] Table 2

[0114]

[0115]

[0116] The UCIQE score was significantly better than all the comparison methods, reaching 0.8893, indicating that the present invention has obvious advantages in color reproduction and saturation control, and is particularly suitable for scenarios with complex water quality and severe degradation.

[0117] To further verify the consistency of color restoration, this invention also conducted statistical analysis on the color proportions of the RGB three channels. The comparison revealed that this invention can restore the proportions of the red, green, and blue channels to a distribution range close to that of a natural image, significantly outperforming traditional methods with a severe blue-green bias, and also exhibiting higher color stability than depth-based methods.

[0118] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. An adaptive underwater image enhancement method based on multi-domain fusion and color compensation, characterized in that, Includes the following steps: S1. Acquire underwater images and perform data preprocessing on the underwater images; S2. Perform adaptive underwater defogging compensation processing on the preprocessed underwater image; S3. Perform color balancing on the underwater image after adaptive underwater defogging compensation to complete underwater image enhancement.

2. The adaptive underwater image enhancement method based on multi-domain fusion and color compensation according to claim 1, characterized in that, S1 includes the following sub-steps: S11. Construct a hybrid multi-scale illumination model and use it to correct the global and local illumination of underwater images to generate enhanced images; S12. Based on the enhanced image, generate a preprocessed underwater image using a wavelet-guided adaptive filtering method.

3. The adaptive underwater image enhancement method based on multi-domain fusion and color compensation according to claim 2, characterized in that, In S11, the enhanced image I hmim The formula for calculating (x,y) is: L(x,y)=α(x,y)·L global (x,y)+(1-α(x,y))·L local (x,y); Where x represents the x-coordinate, y represents the y-coordinate, and L global (x,y) represents the global brightness estimate, L local (x,y) represents the local brightness estimate. I represents the result after applying global gamma correction to the input image. in (x,y) represents the intensity of the original gray value of the input image at coordinates (x,y), and S represents the number of layers in the multi-scale Gaussian filter. The scale is represented by σ. s Gaussian kernel, ω s Let L(x,y) represent the corresponding weight coefficients, L(x,y) represent the final illumination map, α(x,y) represent the adaptive weights at each location in the image, and ε represent a very small positive number.

4. The adaptive underwater image enhancement method based on multi-domain fusion and color compensation according to claim 2, characterized in that, S12 includes the following sub-steps: S121. Use discrete wavelet transform to decompose the enhanced image into several frequency sub-bands; S122. In each frequency sub-band, determine the noise level based on the local variance; S123. Perform bilateral filtering on the low noise level region and high-pass filtering on the high noise level region; wherein, the high noise level region is a local region with a signal-to-noise ratio of less than 20dB, and the low noise level region is a region with a signal-to-noise ratio of 20dB or higher. S124. Combine the filtered frequency subbands to generate a preprocessed underwater image.

5. The adaptive underwater image enhancement method based on multi-domain fusion and color compensation according to claim 4, characterized in that, In S122, the noise level N est The formula for calculating (x,y) is: Where Ω represents the neighborhood window, μ Ω Let (x, y) represent the average value within the window, (x, y) represent the pixel coordinates of the current noise level to be estimated, and i represent the loop variable when traversing all pixels in the neighborhood window. hmin (i) represents the gray value of the i-th pixel within the neighborhood window.

6. The adaptive underwater image enhancement method based on multi-domain fusion and color compensation according to claim 1, characterized in that, S2 includes the following sub-steps: S21. Calculate the spectral attenuation ratio based on the preprocessed underwater image; S22. Calculate the water body type corresponding to the preprocessed underwater image based on the spectral attenuation ratio; S23. Based on the water body type, use the optimization objective function to perform multi-priority fusion compensation to complete the adaptive underwater fogging compensation process.

7. The adaptive underwater image enhancement method based on multi-domain fusion and color compensation according to claim 6, characterized in that, In S21, the spectral attenuation ratio η c The calculation formula is: Where, μ c μ represents the average brightness of a channel in an image. R μ represents the average brightness of channel R in the image. G μ represents the average brightness of channel G in the image. B This represents the average brightness of channel B in the image; In step S22, the formula for calculating the water body type (WTI) corresponding to the preprocessed underwater image is as follows: in, This represents the reference attenuation ratio for the ω-th type of water body; In step S23, the underwater image I after adaptive underwater defogging compensation processing audc The formula for calculating (x) is: Where x represents the horizontal coordinate, J represents the objective function or image to be optimized, J(x) represents the function value or pixel value of the objective function or image at a certain location, and J... j d(x) represents the recovery result obtained from the j-th prior, d(x) represents the image depth, and λ represents the depth of the image. j (d(x)) represents the weighting coefficient determined by the image depth and water body type. β represents the total variation regularization term, and β represents the regularization weight parameter.

8. The adaptive underwater image enhancement method based on multi-domain fusion and color compensation according to claim 1, characterized in that, S3 includes the following sub-steps: S31. Construct a reference color distribution; S32. Compare the underwater image after adaptive underwater defogging compensation with the reference color distribution, and calculate the CIEDE2000 color difference of the image at the pixel level. S33. Based on the CIEDE2000 color difference at the pixel level, perform color correction to complete underwater image enhancement.

9. The adaptive underwater image enhancement method based on multi-domain fusion and color compensation according to claim 8, characterized in that, In step S33, the formula for calculating color correction is as follows: I pdc (x)=I audc (x)+β·(ΔE target (x)-ΔE current (x)); Where x represents the x-coordinate, I pdc (x) represents the final optimized output value, I audc (x) represents the underwater image after adaptive underwater defogging compensation processing, β represents the correction step size, and ΔE target (x) represents the color difference threshold that is difficult for the human eye to distinguish, ΔE current (x) represents the perceptual error between the current pixel and the reference.

Images