Deep sea bottom blurred image enhancement method based on color compensation and image noise reduction

Abstract

The invention discloses a deep sea bottom blurred image enhancement method based on color compensation and image noise reduction, and the method comprises the following steps: designing an automatic white balance algorithm of brightness and color balance, and carrying out the color compensation of an underwater image; a wavelet domain heat conduction matrix is adopted to process the underwater image after white balance, image noise is reduced, and edge details are enhanced; designing a compressed domain boundary averaging algorithm to eliminate the blocking effect of the underwater image; and performing multi-scale image fusion based on the Laplacian pyramid to obtain an enhanced clear underwater image. According to the method, the definition of the underwater image can be effectively improved, the color of the underwater image is recovered, the contrast ratio is high, and the visibility is good; the method can reduce the noise of the image and eliminate the blocking effect of the underwater image, improves the quality of the underwater degraded image, improves the visual effect, and is more suitable for the underwater turbid environment.

Description

Technical Field

[0001] This invention relates to the field of visual perception technology, and in particular to a method for enhancing blurred deep-sea images based on color compensation and image denoising. Background Technology

[0002] Visual perception of the deep-sea environment is a prerequisite for the smooth operation of exploration and sampling vehicles, effectively improving their mobility and drilling sampling convenience. As an important source of marine information from the outside world, high-quality visual images are the foundation for exploration and sampling vehicles to enhance visual perception and perform 3D topographic reconstruction. However, due to the complexity of the deep-sea environment, disturbance of seabed sediments, and the absorption and scattering of light during underwater propagation, visual images are prone to color distortion. At the same time, the irregular movement of suspended particles introduces a large amount of random noise, which significantly affects image quality, making the images blurry and severely impacting the visual performance of exploration and sampling vehicles.

[0003] Currently, many scholars have conducted research on image enhancement methods for underwater turbid scenes. Common physical model-based methods are widely used in image restoration, mainly by estimating model parameters to remove image blur. Common physical model-based methods include UDCP (Underwater Dark Channel Prior) and IBLA (Image Blurriness and Light Absorption). However, due to the complexity of natural and artificial light during real-time underwater shooting, it is difficult to effectively explain the relationship between water absorption of light and color attenuation using physical model-based methods. Moreover, accurately estimating physical model parameters is also a challenge. Therefore, using physical model methods to restore underwater images often introduces more severe color casts and reduces the overall contrast of the image.

[0004] Therefore, researchers began to use pixel intensity redistribution-based methods for image processing. Pixel intensity redistribution-based image enhancement methods produce better visual effects by changing pixel values ​​in the spatial or transform domains. This method does not rely on any model or prior knowledge and has been widely used in the field of underwater image enhancement. However, image enhancement methods often ignore the physical characteristics of underwater light propagation, such as the degree of degradation of underwater images and scene depth information, so they cannot fully recover the color and texture information of underwater images. With the rapid development of deep learning, especially convolutional neural networks (CNNs) and generative adversarial networks (GANs), they have been widely used in underwater image enhancement. However, deep learning-based image enhancement methods require large image datasets in practical applications, increasing time costs and computational resource consumption, and the realism of the generated underwater images is almost impossible to verify. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a simple algorithm with good visual effects for enhancing blurred deep-sea images based on color compensation and image denoising.

[0006] The technical solution of this invention to solve the above-mentioned technical problems is: a method for enhancing blurred deep-sea images based on color compensation and image denoising, comprising the following steps:

[0007] Step 1: Design an automatic white balance algorithm that balances brightness and color to compensate for color in underwater images;

[0008] Step 2: The underwater image after white balance is processed using a wavelet domain heat conduction matrix to reduce image noise and enhance edge details;

[0009] Step 3: Design a compressed domain boundary averaging algorithm to eliminate the block effect in underwater images;

[0010] Step 4: Perform multi-scale image fusion based on the Laplacian pyramid to obtain an enhanced and clear underwater image.

[0011] The above-mentioned method for enhancing blurred deep-sea images based on color compensation and image denoising, specifically step 1, is as follows:

[0012] Step 11: Use the grayscale world white balance algorithm to compensate for image color distortion caused by the selective absorption of light by water at different depths. The calculation formula for red channel compensation in underwater images is as follows:

[0013]

[0014] In the formula, This is the compensation value for the red channel of the image. This is the channel value for the red channel. This is the channel value for the green channel. This represents the average value of the red channel. This represents the average value of the green channel. This is the compensation coefficient for the red channel;

[0015] Step 12: Divide the average grayscale value by the average value of each channel to obtain the weight of each channel. Then multiply each channel by its weight to adjust the grayscale value of each channel. The calculation formula is as follows:

[0016]

[0017]

[0018] In the formula, The average grayscale value. , and These are the weights for the three RGB channels, respectively. This is the channel value of the blue channel. The average value of the blue channel. , and These are the adjusted grayscale values ​​for the three RGB channels;

[0019] Step 13: First, the ambient light is calculated using the white block Retinex algorithm. Then, the light transmittance is estimated using the dark channel prior algorithm. Based on the estimated light transmittance, the white part in the underwater image is determined. Finally, the reference white point is used to adjust the underwater image and correct the color distortion of the image in order to maintain the color constancy of the underwater image.

[0020] Step 14: Determine the white areas by setting a threshold to eliminate highly saturated areas of the image;

[0021] Step 15: After determining the white area, calculate the calibration gain of the three color channels using the average value of the white area, and adjust the underwater image by referring to the white point to obtain the underwater image after white balance.

[0022] In the above-described method for enhancing blurred deep-sea images based on color compensation and image denoising, the formula for determining the white region in step 14 is as follows:

[0023]

[0024] In the formula, This represents the binarized image of the white region after thresholding. To transform the threshold, The average transmittance. express The minimum values ​​of the red, green, and blue channels in the neighborhood. This is a color-corrected image of an underwater scene. Light transmittance, , These represent the horizontal and vertical coordinates of the pixel, respectively.

[0025] In the above-described method for enhancing blurred deep-sea images based on color compensation and image denoising, step 14... The calculation formula is:

[0026]

[0027] In the formula, and These are the width and height of the image, respectively.

[0028] The above-mentioned deep-sea blurred image enhancement method based on color compensation and image denoising, the specific process of step 2 is as follows:

[0029] Step 21: Convert the white balance pre-processed image from the RGB color space to the HSI color space, which is the chroma-saturation-intensity color space, to obtain the I channel image. Then, use the heat conduction matrix HCM to perform a 3×3 mask translation operation on the I channel image.

[0030] Step 22: Perform multi-scale soft thresholding denoising on the I-channel image, using high-pass and low-pass filters for wavelet decomposition;

[0031] First, the high-frequency detail coefficients are enhanced to obtain the enhanced high-frequency detail coefficients, as shown in the formula:

[0032]

[0033] In the formula, For the first Size No. High-frequency detail coefficients in direction, For the enhanced first Size No. High-frequency detail coefficients in direction, For scale indexing, For direction index, For enhancement coefficient;

[0034] Then, the threshold is applied for updating:

[0035]

[0036] In the formula, For the first Size No. Threshold for direction;

[0037] Then, the inverse wavelet transform is calculated and the wavelet coefficients are reconstructed. Finally, the image is returned to the spatial domain through the inverse Fourier transform to obtain the denoised image.

[0038] In the above-mentioned deep-sea blurred image enhancement method based on color compensation and image denoising, the formula for calculating the heat conduction value during mask translation in step 21 is as follows:

[0039]

[0040] In the formula, For the center pixel of the mask thermal conductivity value, It is the average surface area of ​​the thermal paths in the mask. This is the path length from the highest grayscale value to the lowest grayscale value in the mask. The thermal conductivity coefficient, and These are the maximum and minimum gray levels in the mask, respectively.

[0041] In the above-described method for enhancing blurred deep-sea images based on color compensation and image denoising, the formula for calculating the thermal conductivity coefficient in step 21 is as follows:

[0042]

[0043] In the formula, For the center pixel of the mask, For the first Adjacent pixels, The index of the adjacent pixel is used; by comparing the center pixel of the mask with the adjacent pixels, the heat conduction matrix HCM is determined to be positive, negative, or zero.

[0044] The above-mentioned deep-sea blurred image enhancement method based on color compensation and image denoising, the specific process of step 3 is as follows:

[0045] Step 31: Convert the pre-processed underwater image from the RGB color space to the YCbCr color space to obtain the Y component. Then, perform a discrete cosine transform on the Y component, using the idea of ​​image segmentation and classification. Assume the two-dimensional image is... , , , Let be the side length of the sub-image, and then the discrete cosine transform coefficients of an image are obtained as follows:

[0046]

[0047] In the formula, These are the discrete cosine transform coefficients. This represents the position of the discrete cosine transform coefficients. , , The position of the pixel in the original image;

[0048] Step 32: Set the sub-image size to 8×8. Two adjacent sub-images in the horizontal direction that have undergone local homomorphic filtering are... and Through the and Mean filtering is applied to adjacent boundary pixels to eliminate blockiness in the sub-image in the horizontal direction, as shown in the following expression:

[0049]

[0050] In the formula, The width of the template in the horizontal direction. , It is a constant; After processing by the boundary average algorithm Pixel values; After processing by the boundary average algorithm Pixel values; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; , , All are constants;

[0051] Step 33: Two adjacent sub-images in the vertical direction that have undergone local homomorphic filtering are... and Through the and Mean filtering is applied to adjacent boundary pixels to eliminate blockiness in the sub-image in the vertical direction, as shown in the following expression:

[0052]

[0053] In the formula, The vertical template height. , It is a constant; After processing by the boundary average algorithm Pixel values; After processing by the boundary average algorithm Pixel values; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column.

[0054] In the above-mentioned deep-sea blurred image enhancement method based on color compensation and image denoising, in step 31, The value can be:

[0055] .

[0056] In the above-mentioned deep-sea blurred image enhancement method based on color compensation and image denoising, step 4 firstly uses the image after block artifact elimination as the input image, then extracts the dark channel weight, saturation weight, brightness weight, exposure weight, saliency weight and chromaticity weight corresponding to the input image, then calculates the Laplacian value of the input image and the Gaussian value of the corresponding weight mapping, and finally uses multi-scale fusion to obtain the enhanced clear underwater image.

[0057] The beneficial effects of this invention are as follows:

[0058] 1. This invention combines the characteristics of underwater optical imaging and designs an automatic white balance algorithm that balances brightness and color to achieve image color compensation. This can effectively improve the clarity of underwater images, restore the colors of underwater images, and provide high contrast and good visibility.

[0059] 2. This invention utilizes wavelet domain heat conduction matrix to reduce image noise and enhance edge details, and designs a compressed domain boundary averaging algorithm to eliminate image blockiness. This effectively reduces image noise and eliminates blockiness in underwater images, improving the quality of degraded underwater images, enhancing visual effects, and making it more adaptable to turbid underwater environments. It can better provide technical support for high-quality visual perception of deep-sea exploration and sampling vehicles. Attached Figure Description

[0060] Figure 1 This is the overall flowchart of the present invention.

[0061] Figure 2 This is a schematic diagram of the original underwater image.

[0062] Figure 3 This is a schematic diagram of the original underwater image after processing with a grayscale world white balance algorithm.

[0063] Figure 4 A schematic diagram showing the original underwater image after color compensation.

[0064] Figure 5 This is a schematic diagram of an I-channel image.

[0065] Figure 6 This is a schematic diagram of the I-channel image after processing by the heat conduction matrix.

[0066] Figure 7 This is a schematic diagram of the underwater image after noise reduction.

[0067] Figure 8 This is a schematic diagram of converting a denoised underwater image to RGB space.

[0068] Figure 9 This is a schematic diagram of the Y component.

[0069] Figure 10 This is a schematic diagram after performing a discrete cosine transform on the Y component.

[0070] Figure 11 This is a schematic diagram of an underwater image after the block effect has been eliminated.

[0071] Figure 12 A schematic diagram illustrating the conversion of an underwater image to RGB space after eliminating blockiness.

[0072] Figure 13 This is a schematic diagram of the enhanced underwater image obtained using the present invention.

[0073] Figure 14 This is the RGB histogram of the original underwater image.

[0074] Figure 15 This is the RGB histogram of the enhanced underwater image obtained using the method of the present invention. Detailed Implementation

[0075] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0076] like Figure 1 As shown, the deep-sea blurred image enhancement method based on color compensation and image denoising includes the following steps:

[0077] Step 1: Design an automatic white balance algorithm for brightness and color balance to perform color compensation on underwater images. The original underwater image is shown below. Figure 2 As shown.

[0078] The specific process of step 1 is as follows:

[0079] Step 11: Use the grayscale world white balance algorithm to compensate for image color distortion caused by the selective absorption of light by water at different depths, and obtain... Figure 3 The formula for calculating red channel compensation in underwater images is as follows:

[0080]

[0081] In the formula, This is the compensation value for the red channel of the image. This is the channel value for the red channel. This is the channel value for the green channel. This represents the average value of the red channel. This represents the average value of the green channel. This is the compensation coefficient for the red channel.

[0082] Step 12: Divide the average grayscale value by the average value of each channel to obtain the weight of each channel. Then multiply each channel by its weight to adjust the grayscale value of each channel. The calculation formula is as follows:

[0083]

[0084]

[0085] In the formula, The average grayscale value. , and These are the weights for the three RGB channels, respectively. This is the channel value of the blue channel. The average value of the blue channel. , and These are the adjusted grayscale values ​​for the three RGB channels.

[0086] Step 13: First, the ambient light is calculated using the white block Retinex algorithm. Then, the light transmittance is estimated using the dark channel prior algorithm. Next, the white parts in the underwater image are determined based on the estimated light transmittance. Finally, the underwater image is adjusted using a reference white point to correct the color distortion of the image and maintain the color constancy of the underwater image.

[0087] Step 14: Eliminate highly saturated areas in the image by setting a threshold, thereby determining the white areas. The formula for determining the white areas is:

[0088]

[0089] In the formula, This represents the binarized image of the white region after thresholding. To transform the threshold, The average transmittance. express The minimum values ​​of the red, green, and blue channels in the neighborhood. This is a color-corrected image of an underwater scene. Light transmittance, , These represent the horizontal and vertical coordinates of the pixel, respectively.

[0090] The calculation formula is:

[0091]

[0092] In the formula, and These are the width and height of the image, respectively.

[0093] Step 15: After determining the white area, calculate the calibration gain of the three color channels using the average value of the white area. Adjust the underwater image using the reference white point to obtain the white-balanced underwater image, as shown below. Figure 4 As shown.

[0094] Step 2: The underwater image after white balance is processed using a wavelet domain thermal conduction matrix to reduce image noise and enhance edge details.

[0095] The specific process of step 2 is as follows:

[0096] Step 21: Convert the white balance pre-processed image from the RGB color space to the HSI color space. The HSI color space is the chroma-saturation-intensity color space, resulting in the I-channel image. The I-channel image is shown below. Figure 5 As shown; then, a 3×3 mask translation operation is performed on the I-channel image using the heat conduction matrix (HCM) to increase image contrast and detail texture while preserving edge information. The resulting image is shown below. Figure 6 As shown.

[0097] The formula for calculating the heat conduction value during mask translation is:

[0098]

[0099] In the formula, For the center pixel of the mask thermal conductivity value, It is the average surface area of ​​the thermal paths in the mask. This is the path length from the highest grayscale value to the lowest grayscale value in the mask. The thermal conductivity coefficient, and These are the maximum and minimum gray levels in the mask, respectively.

[0100] The formula for calculating the thermal conductivity coefficient is:

[0101]

[0102] In the formula, For the center pixel of the mask, For the first Adjacent pixels, The index of the adjacent pixel is used; by comparing the center pixel of the mask with the adjacent pixels, the heat conduction matrix HCM is determined to be positive, negative, or zero.

[0103] Step 22: Perform multi-scale soft thresholding denoising on the I-channel image, using high-pass and low-pass filters for wavelet decomposition;

[0104] First, the high-frequency detail coefficients are enhanced to obtain the enhanced high-frequency detail coefficients, as shown in the formula:

[0105]

[0106] In the formula, For the first Size No. High-frequency detail coefficients in direction, For the enhanced first Size No. High-frequency detail coefficients in direction, For scale indexing, For direction index, For enhancement coefficient;

[0107] Then, the threshold is applied for updating:

[0108]

[0109] In the formula, For the first Size No. Threshold for direction;

[0110] The purpose of multi-scale soft thresholding is: large coefficients leave a margin of error → soft thresholding; small coefficients are directly set to zero → noise reduction; individual thresholding for each scale → multi-scale adaptation.

[0111] Then, the inverse wavelet transform is calculated and the wavelet coefficients are reconstructed. Finally, the image is returned to the spatial domain via the inverse Fourier transform to obtain the denoised image, such as... Figure 7 As shown; the final image converted to RGB space is as follows. Figure 8 As shown.

[0112] Step 3: Design a compression domain boundary averaging algorithm to eliminate the block effect in underwater images.

[0113] The specific process of step 3 is as follows:

[0114] Step 31: Convert the pre-processed underwater image from the RGB color space to the YCbCr color space to obtain the Y component, as shown below. Figure 9 As shown; then perform a discrete cosine transform operation on the Y component, as follows. Figure 10 As shown, the idea of ​​image segmentation and classification is adopted, assuming that the two-dimensional image is... , , , Let be the side length of the sub-image, and then the discrete cosine transform coefficients of an image are obtained as follows:

[0115]

[0116] In the formula, These are the discrete cosine transform coefficients. This represents the position of the discrete cosine transform coefficients. , , The position of the pixel in the original image; The value can be:

[0117] .

[0118] Step 32: Set the sub-image size to 8×8. Two adjacent sub-images in the horizontal direction that have undergone local homomorphic filtering are... and Through the and Mean filtering is applied to adjacent boundary pixels to eliminate blockiness in the sub-image in the horizontal direction, as shown in the following expression:

[0119]

[0120] In the formula, The width of the template in the horizontal direction. , It is a constant; After processing by the boundary average algorithm Pixel values; After processing by the boundary average algorithm Pixel values; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; , , All are constants;

[0121] Step 33: Two adjacent sub-images in the vertical direction that have undergone local homomorphic filtering are... and Through the and Mean filtering is applied to adjacent boundary pixels to eliminate blockiness in the sub-image in the vertical direction, as shown in the following expression:

[0122]

[0123] In the formula, The vertical template height. , It is a constant; After processing by the boundary average algorithm Pixel values; After processing by the boundary average algorithm Pixel values; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column.

[0124] Underwater images after block effect removal, as shown Figure 11 As shown, the underwater image after blockiness removal is converted to RGB space as follows: Figure 12 As shown.

[0125] Step 4: Perform multi-scale image fusion based on the Laplacian pyramid to obtain an enhanced and clear underwater image.

[0126] In step 4, the image after blockiness removal is first used as the input image. Then, the dark channel weights, saturation weights, brightness weights, exposure weights, saliency weights, and chroma weights corresponding to the input image are extracted. Next, the Laplacian of the input image and the Gaussian values ​​of the corresponding weight mappings are calculated. Finally, multi-scale fusion is used to obtain the enhanced underwater image, such as... Figure 13 As shown.

[0127] Figure 14 This is the RGB histogram of the original underwater image. Figure 15 The RGB histogram of the enhanced underwater image obtained using the method of this invention is based on... Figure 14 and Figure 15It can be seen that the R, G, and B components of the original image are concentrated, the distribution of the three channels is severely unbalanced, the red channel is severely missing, resulting in a clear blue-green bias in the image; the peak values ​​of each channel are concentrated in the low-to-medium intensity region, and the pixel values ​​are densely distributed, resulting in low overall image contrast and blurred details.

[0128] Images processed using the method proposed in this invention exhibit a wider and more uniform histogram distribution. The three-channel distribution is more balanced, and color cast is significantly corrected. The wider pixel value distribution effectively enhances image contrast and detail visibility.

[0129] Algorithm Comparison:

[0130] The method of this invention is compared with six other methods, namely: UDCP (Underwater Dark Channel Prior); IBLA (Underwater Image Restoration Based on Image Blur and Light Absorption); HWD (Histogram-based Underwater Image Dehazing); WCID (Wavelength Compensation and Image Dehazing); FUSION (Multi-scale Image Fusion); and ARC (Adaptive Retinex Color Correction).

[0131] To verify the effectiveness of this invention in image restoration in turbid underwater scenes, qualitative and quantitative comparisons were conducted. Qualitative comparisons analyzed aspects such as image contrast, visibility, and color restoration to explore the processing effects of each method. However, due to differences in individual subjective perception, conclusions may vary. Therefore, to avoid bias in qualitative comparisons, quantitative comparisons were performed, objectively evaluating the restoration effects of different algorithms on underwater images from aspects such as color restoration, contrast, and overall performance. Four main evaluation metrics were used: Information Entropy (IE), Average Gradient (AVG), Underwater Color Image Quality Evaluation (UCIQE), and Naturalness Evaluation (NIQE).

[0132] Information entropy (IE) reflects the average information content level of an image. Generally, a higher value indicates richer information content and higher image fidelity. Average gradient (AVG) reflects the rate of change of image details, representing the image's sharpness. Underwater color image quality assessment index (UCIQE) uses chroma, saturation, and contrast as measurement components, linearly superimposing them. It is mainly used to quantify the degree of image degradation caused by non-uniform lighting, color deviation, turbidity, and blurring in underwater images.

[0133]

[0134] Table 1 shows the quantitative comparison results of different algorithms in underwater turbidity scenarios. The bolded values ​​in the table represent the optimal values ​​of the evaluation indicators. As can be seen from Table 1, the information entropy (IE) of the method of this invention is 7.7972, the average gradient (AVG) is 3.9227, the underwater color image quality evaluation (UCIQE) is 0.6534, and the naturalness evaluation (NIQE) is 3.5229. The method outperforms the other six classic algorithms in IE, AVG, and UCIQE, indicating that the enhanced underwater image has rich information, high clarity, and contrast that is closer to the natural scene, resulting in better visual effects.

[0135] To verify the effectiveness of the method of this invention in restoring turbid underwater images on publicly available underwater image datasets, the publicly available underwater images were obtained from the OCEANDARK and UIEBD image datasets. These datasets provide realistic underwater scenes, including non-uniform low-light conditions, dark area artifacts, underwater color cast environments, and underwater images with blur and noise. Eighty-five turbid and blurred underwater images were randomly selected for experiments, and quantitative data of different image enhancement methods on publicly available underwater image datasets were compared. The information entropy (IE), average gradient (AVG), underwater color image quality assessment (UCIQE), and image naturalness assessment (NIQE) were calculated for each image.

[0136]

[0137] Table 2 shows the quantitative comparison results of different methods on the publicly available underwater image dataset. As can be seen from Table 2, the information entropy of the method of this invention is 7.7957, the average gradient is 13.8040, the underwater color image quality evaluation is 0.6024, and the image naturalness evaluation is 3.5283. The method outperforms the other six classic algorithms in IE, AVG, and UCIQE. The NIQE index value is slightly lower, but the overall effect is better.

Claims

1. A method for enhancing blurred deep-sea images based on color compensation and image denoising, characterized in that, Includes the following steps: Step 1: Design an automatic white balance algorithm that balances brightness and color to compensate for color in underwater images; Step 2: The underwater image after white balance is processed using a wavelet domain heat conduction matrix to reduce image noise and enhance edge details; Step 3: Design a compressed domain boundary averaging algorithm to eliminate the block effect in underwater images; Step 4: Perform multi-scale image fusion based on the Laplacian pyramid to obtain an enhanced and clear underwater image.

2. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 1, characterized in that, The specific process of step 1 is as follows: Step 11: Use the grayscale world white balance algorithm to compensate for image color distortion caused by the selective absorption of light by water at different depths. The calculation formula for red channel compensation in underwater images is as follows: ； In the formula, This is the compensation value for the red channel of the image. This is the channel value for the red channel. This is the channel value for the green channel. This represents the average value of the red channel. This represents the average value of the green channel. This is the compensation coefficient for the red channel; Step 12: Divide the average grayscale value by the average value of each channel to obtain the weight of each channel. Then multiply each channel by its weight to adjust the grayscale value of each channel. The calculation formula is as follows: ； ； In the formula, The average grayscale value. , and These are the weights for the three RGB channels, respectively. This is the channel value of the blue channel. The average value of the blue channel. , and These are the adjusted grayscale values ​​for the three RGB channels; Step 13: First, the ambient light is calculated using the white block Retinex algorithm. Then, the light transmittance is estimated using the dark channel prior algorithm. Based on the estimated light transmittance, the white part in the underwater image is determined. Finally, the reference white point is used to adjust the underwater image and correct the color distortion of the image in order to maintain the color constancy of the underwater image. Step 14: Determine the white areas by setting a threshold to eliminate highly saturated areas of the image; Step 15: After determining the white area, calculate the calibration gain of the three color channels using the average value of the white area, and adjust the underwater image by referring to the white point to obtain the underwater image after white balance.

3. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 2, characterized in that, In step 14, the formula for determining the white area is: ； In the formula, This represents the binarized image of the white region after thresholding. To transform the threshold, The average transmittance. express The minimum values ​​of the red, green, and blue channels in the neighborhood. This is a color-corrected image of an underwater scene. Light transmittance, , These represent the horizontal and vertical coordinates of the pixel, respectively.

4. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 3, characterized in that, In step 14, The calculation formula is: ； In the formula, and These are the width and height of the image, respectively.

5. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 4, characterized in that, The specific process of step 2 is as follows: Step 21: Convert the white balance pre-processed image from the RGB color space to the HSI color space, which is the chroma-saturation-intensity color space, to obtain the I channel image. Then, use the heat conduction matrix HCM to perform a 3×3 mask translation operation on the I channel image. Step 22: Perform multi-scale soft thresholding denoising on the I-channel image, using high-pass and low-pass filters for wavelet decomposition; First, the high-frequency detail coefficients are enhanced to obtain the enhanced high-frequency detail coefficients, as shown in the formula: ； In the formula, For the first Size High-frequency detail coefficients in direction, For the enhanced first Size High-frequency detail coefficients in direction, For scale indexing, For direction index, For enhancement coefficient; Then, the threshold is applied for updating: ； In the formula, For the first Size Threshold for direction; Then, the inverse wavelet transform is calculated and the wavelet coefficients are reconstructed. Finally, the image is returned to the spatial domain through the inverse Fourier transform to obtain the denoised image.

6. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 5, characterized in that, In step 21, the formula for calculating the heat conduction value during mask translation is: ； In the formula, For the center pixel of the mask thermal conductivity value, It is the average surface area of ​​the thermal paths in the mask. This is the path length from the highest grayscale value to the lowest grayscale value in the mask. The thermal conductivity coefficient, and These are the maximum and minimum gray levels in the mask, respectively.

7. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 6, characterized in that, In step 21, the formula for calculating the thermal conductivity coefficient is: ； In the formula, For the center pixel of the mask, For the first Adjacent pixels, The index of the adjacent pixel is used; by comparing the center pixel of the mask with the adjacent pixels, the heat conduction matrix HCM is determined to be positive, negative, or zero.

8. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 7, characterized in that, The specific process of step 3 is as follows: Step 31: Convert the pre-processed underwater image from the RGB color space to the YCbCr color space to obtain the Y component. Then, perform a discrete cosine transform on the Y component, using the idea of ​​image segmentation and classification. Assume the two-dimensional image is... , , These are the x and y coordinates of the two-dimensional image, respectively. , , Let be the side length of the sub-image, and then the discrete cosine transform coefficients of an image are obtained as follows: ； In the formula, These are the discrete cosine transform coefficients. This represents the position of the discrete cosine transform coefficients. , These are the frequencies in the vertical and horizontal directions, respectively. , , These are the normalization coefficients; Step 32: Set the sub-image size to 8×8. Two adjacent sub-images in the horizontal direction that have undergone local homomorphic filtering are... and Through the and Mean filtering is applied to adjacent boundary pixels to eliminate blockiness in the sub-image in the horizontal direction, as shown in the following expression: ； In the formula, The width of the template in the horizontal direction. , It is a constant; After processing by the boundary average algorithm Pixel values; After processing by the boundary average algorithm Pixel values; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; , , All are constants; Step 33: Two adjacent sub-images in the vertical direction that have undergone local homomorphic filtering are... and Through the and Mean filtering is applied to adjacent boundary pixels to eliminate blockiness in the sub-image in the vertical direction, as shown in the following expression: ； In the formula, The vertical template height. , It is a constant; After processing by the boundary average algorithm Pixel values; After processing by the boundary average algorithm Pixel values; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column; for The Middle Line number The pixel values ​​of the column.

9. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 8, characterized in that, In step 31 The value can be: 。 10. The deep-sea blurred image enhancement method based on color compensation and image denoising according to claim 8, characterized in that, In step 4, the image after block artifact removal is first used as the input image. Then, the dark channel weight, saturation weight, brightness weight, exposure weight, saliency weight, and chroma weight corresponding to the input image are extracted respectively. Next, the Laplacian value of the input image and the Gaussian value of the corresponding weight mapping are calculated. Finally, multi-scale fusion is used to obtain the enhanced clear underwater image.

Images