An underwater dark image processing method based on color space conversion

Abstract

The application discloses a dark underwater image processing method based on color space conversion, which comprises the following steps: firstly, pre-acquired underwater images are down-sampled to reduce the image size and copied into two images; secondly, color space conversion is performed on the two parts of underwater images respectively; thirdly, an underwater optical imaging image model is established, and the first converted underwater image is processed by using an image deblurring algorithm to obtain a clear image with high contrast; fourthly, an improved white balance algorithm is used to process the deblurred image; fifthly, a nonlinear channel prior is introduced into a traditional image deblurring model to minimize the item to eliminate the motion blur problem caused by water flow fluctuation; sixthly, the obtained accurate blur kernel and potential clear image are iterated with the initial restored image to obtain a final restored image; and finally, the two images are fused, re-sampled and outputted to obtain a final clear underwater image. The application can better realize image restoration, improve the brightness and contrast of the image and improve the recognition degree.

Description

Technical Field

[0001] This invention belongs to the field of image processing, specifically relating to a method for processing underwater dark images based on color space conversion. Background Technology

[0002] In recent years, energy issues have become increasingly serious. Despite having one of the world's largest land areas, my country suffers from a severe shortage of many important energy sources. Therefore, exploring underwater environments, including the ocean, has become crucial in the search for more efficient and clean energy sources. With the increasing demand for underwater environment exploration, underwater imaging technology has developed rapidly and is widely used in underwater resource exploration, marine military applications, and underwater robotics research. The underwater environment is completely different from the terrestrial environment. As depth increases, the underwater environment becomes increasingly dark, and particulate matter in the water severely affects image clarity, causing color distortion, low contrast, and blurring, hindering the utilization and analysis of underwater images. Therefore, research on underwater image restoration, deblurring, enhancing image clarity, reconstructing the underwater environment, and constructing underwater environment models are of great significance to marine exploration technology. While numerous methods have emerged in underwater image processing, they often suffer from insufficient computational accuracy, complex and difficult-to-apply algorithms, and unsatisfactory processing results. Researchers in this field are continuously improving computational methods to address these issues. Improving the accuracy of calculations, enhancing image processing performance, and simplifying the image processing process are all challenging aspects of this research and are the key issues addressed in this invention. Summary of the Invention

[0003] Purpose of the invention: To address the problems of dark colors and insufficient brightness in underwater images, this invention provides a method for processing dark underwater images based on color space conversion, which improves the effect and calculation accuracy of underwater image processing, resulting in clearer and more practical underwater images.

[0004] Technical Solution: This invention provides a method for processing underwater low-light images based on color space conversion, comprising the following steps:

[0005] (1) The pre-acquired underwater images are downsampled to reduce the image size and then copied into two images;

[0006] (2) The two underwater images were converted to color space respectively. The RGB color space was converted to the HSV color space, with the H component representing hue, the S component representing saturation, and the V component representing brightness. The V component of the first image was enhanced. The S and V components of the second image were adjusted to increase saturation.

[0007] (3) Establish an underwater optical imaging image model, and use an image deblurring algorithm to process the first converted underwater image to obtain a clear image with high contrast.

[0008] (4) An improved white balance algorithm is used to process the deblurred image;

[0009] (5) To address the underwater image blurring problem caused by forward scattering components, nonlinear channel priors are embedded into the traditional image deblurring model, and this term is minimized to eliminate motion blur caused by water flow fluctuations. The optimal solution of the blur kernel is obtained by using image gradient and fast Fourier transform. The obtained accurate blur kernel and potential clear image are iterated with the initial restored image to obtain the final restored image.

[0010] (6) The first image processed by steps (2) to (5) and the second image processed by step (2) are fused together, resampled, and the final clear underwater image is output.

[0011] Furthermore, the underwater optical imaging image model established in step (3) is as follows:

[0012] I(x,y)=J(x,y)t(x,y)+A[1-t(x,y)] (7)

[0013] Where I(x,y) is the original input image, J(x,y) is the processed deblurred image, A is the underwater light intensity, and t(x,y) is the water transmittance. J(x,y) is calculated according to this formula to obtain a clear underwater image. According to the dark primary color theory, the key to underwater image deblurring lies in solving the water light intensity and water transmittance.

[0014] Furthermore, the light intensity of the water body is estimated using histogram statistical characteristics, as follows:

[0015] The statistical average of the N pixels with the largest grayscale values ​​(0.05%) in the dark primary color image of the underwater image is taken as the water body light intensity value; the statistical average of the dark primary color values ​​of the original image is calculated as the water body illumination intensity value A.

[0016]

[0017] Where h is the horizontal axis value of the histogram; N is the total number of pixels in the dark primary color image.

[0018] Furthermore, the process for calculating the water body transmittance is as follows:

[0019] Keeping the pixel values ​​of the blue and green channels unchanged, invert the pixel values ​​of the red channel to obtain the inverse red channel, and then calculate the transmittance; the formula for calculating the transmittance using the inverse channel is:

[0020]

[0021] By inverting the values ​​of each low-valued channel and correcting for negative values, the improved transmittance t(x, y) is obtained as follows:

[0022]

[0023] Where A(μ) is the underwater light intensity, I c (y, μ) are the color channels of the image before processing, and ω is the correction parameter. Further, the implementation process of step (4) is as follows:

[0024] For underwater images containing a small number of pixels with values ​​reaching the limits of 0 or 255, before performing the affine transformation, a cumulative histogram of pixel values ​​is calculated. Pixels in the histogram that are close to the limits of 0 or 255 are removed, and the maximum value F of the remaining pixels is calculated. max and minimum value F min Then divide the interval [F] min F max The pixel values ​​within the range are stretched to [0, 255] through radial transformation to obtain a clear image after underwater blurring.

[0025] Furthermore, the implementation process of step (5) is as follows:

[0026] By embedding the nonlinear channel prior into the traditional image deblurring model, the objective function is:

[0027]

[0028] Where a, b, and c are the corresponding weight parameters of the regularization term; For gradient operator; L is the potential sharp image; K is the blur kernel; C nl (L) represents the dark channel C of a potentially sharp image. d (L) and the bright channel C of the potentially sharp image b (L) ratio; J c The first constraint term uses the L2 norm to enhance the convolution result of the potentially sharp image L and the blur kernel K with the dehazed image J. c The first constraint term is the similarity of the two terms; the second constraint term is the correlation term of the nonlinear channel prior; the third constraint term is used to preserve the edge information of the image while smoothing out small details; the fourth term is the regularization term on K, which uses the traditional L2 norm to constrain the smoothness of the blur kernel.

[0029] Applying constraints to K yields the following constraints:

[0030]

[0031]

[0032] To improve the accuracy of the blur kernel calculation, K is estimated using the image gradient instead of the intensity:

[0033]

[0034] The optimal solution for K is obtained using the Fast Fourier Transform:

[0035]

[0036] Among them, C d (L) and C b (L) represent the dark and bright channels of the potentially sharp image, respectively; the initial restored image I is obtained using the obtained potentially sharp image L and the calculated precise blur kernel K. i .

[0037] Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are: the present invention simplifies the calculation process while ensuring the clarity of underwater images, making image processing more convenient and faster. At the same time, considering the blurring problem caused by forward scattering components, it further improves the clarity of the image, thereby improving the image recognition. It can be used for rapid processing of underwater images, improving the effect and calculation accuracy of underwater image processing, and obtaining clearer and more practical underwater images. Attached Figure Description

[0038] Figure 1 This is the overall flowchart of image processing in this method;

[0039] Figure 2 Flowchart of the improved underwater image processing algorithm. Detailed Implementation

[0040] The present invention will now be described in further detail with reference to the accompanying drawings.

[0041] This invention proposes a method for processing underwater dark images based on color space conversion, such as... Figure 1 As shown, it includes the following steps:

[0042] Step 1: Downsample the pre-acquired underwater image to reduce its size and copy it into two images; convert the color space of the two underwater images from RGB to HSV, where H is the hue, S is the saturation, and V is the brightness; enhance the V component of the first image; adjust the S and V components of the second image to increase saturation.

[0043] RGB is a commonly used color space. Any color can be created by mixing R (red), G (green), and B (blue) in different proportions. The HSV color space consists of H (hue), S (saturation), and V (lightness / darkness), and is a color space created based on the intuitive characteristics of color. RGB to HSV conversion formula:

[0044] max = max{R, G, B} (1)

[0045] min=min{R,G,B} (2)

[0046] V = max (3)

[0047]

[0048]

[0049] Underwater images in deep water environments are generally dark. After color space conversion, the S and V components in the HSV color space need to be adjusted to improve the brightness and saturation of the output image.

[0050] HSV color space processing: To reduce algorithm complexity, the underwater image is downsampled and then copied twice. Both images undergo color space conversion from RGB to HSV. In the first image, the H and S components remain unchanged, while the V component is increased by a factor of 1.2. This prevents parts of the image from becoming too dark and indistinguishable, while also improving the overall brightness of the final output image. In the second image, the H component remains unchanged, the V component is adaptively decreased, and the S component is linearly stretched. The specific operation of the linear stretching of the S component is as follows:

[0051]

[0052] In the formula, S represents the color space saturation after downsampling. Saturation is represented by denoted as linearly processed , and M is the average pixel value of the downsampled image. After stretching, the image colors are more saturated, resulting in a better visual effect. After processing, the first and second parts of the image are converted to the RGB color space for subsequent deblurring.

[0053] Step 3: Establish an underwater optical imaging model. Process the first converted underwater image using an image deblurring algorithm to obtain a clear image with high contrast, such as... Figure 2 As shown, the process is as follows:

[0054] Ignoring the influence of forward scattering components, underwater optical imaging model:

[0055] I(x,y)=J(x,y)t(x,y)+A[1-t(x,y)] (7)

[0056] Where I(x, y) is the original input image, J(x, y) is the processed deblurred image, A is the underwater light intensity, and t(x, y) is the water transmittance. A clear underwater image is obtained by calculating J(x, y) using this formula.

[0057]

[0058] Among them, J c (x, y) represents the color channel of J(x, y), Ω(x, y) is a local region of the image, and J dark (x, y) is the dark primary color of J(x, y).

[0059] According to the dark primary color theory, the dark primary color value can be approximately equivalent to taking the minimum brightness of each pixel in the RGB channels of an underwater image, that is:

[0060]

[0061] In the formula, c represents the three channels of the original image. It takes the minimum brightness value of the three color channels in the image before processing.

[0062] Assuming the underwater light intensity A is known, estimate the water transmittance and take the minimum value of (7), that is:

[0063]

[0064] According to the dark primary color prior theory, the dark primary color value of the deblurred underwater image tends to zero, that is:

[0065]

[0066] At this point, equation (10) can be transformed into:

[0067]

[0068] Dividing both sides of equation (12) by A, we get:

[0069]

[0070] Underwater transmittance is:

[0071]

[0072] Due to various environmental conditions, a correction factor ω should be added when calculating underwater transmittance.

[0073]

[0074] Substituting into equation (9), equation (15) can be transformed into:

[0075]

[0076] Substituting into equation (7), we can obtain the deblurred image. Simultaneously, setting the transmittance to a lower limit of 0.2, we get:

[0077]

[0078] The underwater optical imaging model was established. Based on the dark primary color theory, the key to deblurring underwater images lies in solving the water light intensity and water transmittance.

[0079] Estimating water body light intensity using histogram statistical properties: Calculations are performed using prior knowledge of dark primary colors, requiring the assumption that the intensity value A of water body illumination is a given value and that the transmittance of the water body is constant within each local region of Ω(x,y).

[0080] To prevent the calculated water light intensity from deviating from reality by taking the pixel with the highest brightness as an object in the water instead of the water body, the statistical average of the 0.05%N (N is the total number of pixels in the dark primary color image) of the underwater image is taken as the water light intensity value.

[0081] Calculate I dark The histogram (hist) of (x, y) is calculated, and then the histograms are summed from right to left to obtain hisstsum. The abscissa h of the histogram is obtained where hisstsum is greater than 0.05%N (N is the total number of pixels in the dark primary color image), and I is calculated in the interval [h-255]. dark The statistical average value of pixels (x, y), i.e., the light intensity of the water body, is:

[0082]

[0083] In the formula, h is the horizontal axis value of the histogram, and N is the total number of pixels in the dark primary color image.

[0084] Improved Transmittance Estimation: Most underwater images are surrounded by a blue-green atmosphere, making them prone to blue-green bias. This bias is primarily due to the significantly different absorption characteristics of water for different wavelengths of light. Water absorbs less light near the blue-green band but absorbs more light near the red band, resulting in severe attenuation of red light in the water. This leads to consistently low pixel values ​​in the red channel after imaging. To address this, we keep the pixel values ​​of the blue-green channels constant and invert the pixel values ​​of the red channel to obtain the inverse red channel. Then, we calculate the transmittance, thus mitigating the impact of severe red light attenuation. The formula for calculating transmittance using the inverse channel is:

[0085]

[0086] Due to the complex underwater environment, besides the blue-green tint, other color shifts or overall darkness may occur, primarily due to low pixel values ​​in certain color channels. Therefore, a reverse pass is performed on each low-valued channel to correct potential negative values, resulting in an improved transmittance formula t(x, y):

[0087]

[0088] In the formula, A(μ) is the underwater light intensity, and I c (y, μ) are the color channels of the image before processing, and ω is the correction parameter.

[0089] Step 4: Process the deblurred image using an improved white balance algorithm.

[0090] Different wavelengths of optical fiber attenuate at varying degrees in water, causing color distortion in underwater images. Furthermore, while removing scattering components improves the clarity of underwater images, the overall brightness remains low, and the contrast is not high. To address these issues, an improved white balance algorithm is used for further processing of the deblurred image.

[0091] Assuming that the maximum pixel value in the RGB channels of an image corresponds to the white area, and the minimum pixel value corresponds to the black area, based on this assumption, and building upon the white balance algorithm, affine transformations are performed on the pixel values ​​of each channel of the input image to stretch the pixel values ​​of the three RGB channels of the underwater image, making them occupy as much of the [0.255] range as possible. This achieves image color correction while improving contrast and enhancing the visual effect of the underwater image.

[0092] For underwater images containing a small number of pixels with values ​​reaching the limits of 0 or 255, achieving color balance using this white balance algorithm is difficult. To ensure the effectiveness of the white balance, before performing the affine transformation, a cumulative histogram of pixel values ​​is calculated. Pixels close to the limits of 0 or 255 in the histogram are removed, and the maximum value F of the remaining pixels is calculated. max and minimum value F min Then divide the interval [F] min F max The pixel values ​​within the range are stretched to [0, 255] through radial transformation to obtain the final clear image after underwater blurring.

[0093] After white balance is applied, the overall brightness and contrast of the image will be improved, severely degraded channels will be improved, and the image will present a better visual effect.

[0094] Step 5: To address the underwater image blurring problem caused by forward scattering components, a nonlinear channel prior is introduced into the traditional image deblurring model. This term is minimized to eliminate motion blur caused by water flow fluctuations. The optimal solution of the blur kernel is obtained using image gradient and fast Fourier transform. The obtained accurate blur kernel and the potentially sharp image are iterated with the initial restored image to obtain the final restored image.

[0095] The dark channel at pixel i in a color image M is defined as:

[0096]

[0097] Where θ1 represents a local image centered at pixel i; the dark channel is the minimum pixel value in pixel module θ1; L c Let M represent the color channels of the color image. Conversely, the bright channels represent the maximum values ​​in a local image block θ2, defined as:

[0098]

[0099] As an image gradually blurs, the dark (light) pixels are averaged with their neighboring larger (smaller) pixels. This causes the dark channel of the blurred image to become larger and the light channel to become smaller. This can be summarized by the following two characteristics:

[0100] C d (B) i ≥C d (L) i (twenty three)

[0101] C b (B) i ≤C b (L) i (twenty four)

[0102] Among them, C d (B) i and C d (L) i These represent the dark channel of the blurred image and the dark channel of the sharp image, respectively; C b (B) i and C b (L) i These represent the bright channels of the blurred image and the bright channels of the sharp image, respectively.

[0103] Because dark and bright channels can distinguish between sharp and blurry images, the ratio of dark to bright channels is defined as a non-linear channel to enhance this ability:

[0104]

[0105] In this case, pixel block θ1 and pixel block θ2 are set to the same size. When C d (M) i When the value is not 0, we can deduce that:

[0106]

[0107] The L1 norm is used to accumulate nonlinear channel elements, and the nonlinear channel prior is defined as follows:

[0108] ψ(L)=||C nl (L)||1 (27)

[0109] Because of C nl (B) i ≥C nl (L) i The following inequality is derived:

[0110] ||C nl (B)||1>||C nl (L)||1. (28)

[0111] Image restoration algorithm based on nonlinear channel prior: Motion blur caused by unstable water flow and blur caused by light diffusion in the underwater scene are unavoidable during underwater imaging. Therefore, it is necessary to consider the processing of forward scattering, which is expressed as follows in the underwater optical imaging model:

[0112] I1(x)=[J(x)t(x)]*K+N(x) (29)

[0113] By processing the forward scattering component, the algorithm adds image blur removal functionality while maintaining dehazing and color correction performance, enabling more comprehensive and realistic recovery of underwater scene images. Simplifying J(x)t(x) to L yields the traditional image deblurring model:

[0114] J c =L*K+N (30)

[0115] Among them, J c L represents the image to be processed after dehazing; K represents the blur kernel; and N represents additive noise.

[0116] Introducing nonlinear channel priors into the traditional image deblurring model, the objective function is:

[0117]

[0118] Where a, b, and c are the corresponding weight parameters of the regularization terms in the above formula; The gradient operator; the first constraint term utilizes the L2 norm to enhance the convolution result of L (potentially sharp image) and K (blur kernel) with the dehazed image J. c The first constraint term is the similarity of the two terms; the second constraint term is the correlation term of the nonlinear channel prior; the third constraint term is used to preserve the edge information of the image while smoothing out small details; the fourth term is the regularization term on K (fuzz kernel). For ease of calculation, the traditional L2 norm is used to constrain the smoothness of the fuzz kernel.

[0119] Since it is impossible to simultaneously determine the two variables L and K, an alternating projection minimization algorithm is used to estimate them alternately. Then, constraints are applied to K, resulting in the following constraints:

[0120]

[0121]

[0122] Among them, C d (L) and C b (L) represent the dark and bright channels of the potentially sharp image, respectively.

[0123] Since equation (31) has no closed-form solution, we use an approximation strategy with C. b (L p ) to replace C b (L):

[0124]

[0125] Among them, C b (L p ) is C b (L) approximate value.

[0126] Equation (33) is optimized using the semi-quadratic splitting method, introducing auxiliary variables u and g:

[0127]

[0128]

[0129] Where λ1 and λ2 are penalty term parameters; when the values ​​of λ1 and λ2 are infinitely close, the solution of equation (34) is equivalent to the solution of equation (33); by fixing other values ​​and alternately optimizing L, u, and g, equation (34) is solved in two steps:

[0130] First, fix L to calculate u and g:

[0131]

[0132]

[0133] The second step is to fix u and g to calculate L:

[0134]

[0135] To reduce the solution complexity, equation (38) is transformed into the frequency domain and written as a vector matrix multiplication:

[0136]

[0137] Where K, u, and g are the vector matrix forms of K, u, and g, respectively.

[0138] Since equation (39) is a least squares problem, it cannot be solved directly using the fast Fourier transform. Therefore, an auxiliary variable v is introduced in the second constraint term L:

[0139]

[0140] Where λ3 is a positive penalty term; by solving L and v in equation (40) using the iterative method, the following constraints are obtained;

[0141]

[0142]

[0143] Equations (41) and (42) both have closed-form solutions; Equation (41) is solved using Fast Fourier Transform:

[0144]

[0145] Where L represents the potentially sharp image. F -1 (·)and represents the Fourier transform, inverse Fourier transform, and complex conjugate operator, respectively; ⊙ is the element-wise multiplication symbol; subscripts x and y represent the partial derivatives of each quantity in the x-direction and y-direction, respectively.

[0146] To improve the accuracy of the blur kernel calculation, K is estimated using the image gradient instead of the intensity in equation (33):

[0147]

[0148] The optimal solution for K is obtained using the Fast Fourier Transform:

[0149]

[0150] The initial restored image I is obtained by using the potentially sharp image L and the precisely blurred kernel K. i For the potentially sharp image L and the initial restored image I iAfter multiple iterations, if the number of iterations is less than 4, return to equation (43) to iterate and solve for the potential sharp image L until the number of iterations is 4. Then, for the initial restored image I... i The fuzzy kernel K is iterated multiple times. If the number of iterations is less than 10, the equation (45) is returned to iterate the fuzzy kernel K until the number of iterations is 10.

[0151] Step 6: Fuse the first image processed in steps 2 to 5 with the second image processed in step 2, resample, and output the final clear underwater image.

[0152] The two processed images are then fused using the following formula:

[0153] T(x)=0.7F1(x)+0.3F2(x)+0.5 (46)

[0154] Where F1(x) and F2(x) are the first and second images before fusion, respectively. The two images are weighted and fused to obtain the enhanced output image.

Claims

1. A method for processing underwater low-light images based on color space conversion, characterized in that, Includes the following steps: (1) The pre-acquired underwater images are downsampled to reduce the image size and then copied into two images; (2) The two underwater images were converted to color space respectively. The RGB color space was converted to the HSV color space, with the H component representing hue, the S component representing saturation, and the V component representing brightness. The V component of the first image was enhanced. The S and V components of the second image were adjusted to increase saturation. (3) Establish an underwater optical imaging image model, and use an image deblurring algorithm to process the first converted underwater image to obtain a clear image with high contrast. (4) An improved white balance algorithm is used to process the deblurred image; (5) To address the underwater image blurring problem caused by forward scattering components, nonlinear channel priors are embedded into the traditional image deblurring model, and this term is minimized to eliminate motion blur caused by water flow fluctuations. The optimal solution of the blur kernel is obtained by using image gradient and fast Fourier transform. The obtained accurate blur kernel and potential clear image are iterated with the initial restored image to obtain the final restored image. (6) The first image processed by steps (2) to (5) and the second image processed by step (2) are fused together, resampled and the final clear underwater image is output; The underwater optical imaging image model established in step (3) is as follows: I(x,y)=J(x,y)t(x,y)+A[1-t(x,y)] (7) Where I(x,y) is the original input image, J(x,y) is the processed deblurred image, A is the underwater light intensity, and t(x,y) is the water transmittance. J(x,y) is calculated according to this formula to obtain a clear underwater image. According to the dark primary color theory, the key to underwater image deblurring lies in solving the water light intensity and water transmittance. The process for calculating the water body transmittance is as follows: Keeping the pixel values ​​of the blue and green channels unchanged, invert the pixel values ​​of the red channel to obtain the inverse red channel, and then calculate the transmittance; the formula for calculating the transmittance using the inverse channel is: By inverting the values ​​of each low-valued channel and correcting for negative values, the improved transmittance t(x,y) is obtained as follows: where A(μ) is the underwater light intensity, I c (y, μ) is the color channel of the pre-processed image, and ω is a correction parameter.

2. The underwater low-light image processing method based on color space conversion according to claim 1, characterized in that, The implementation process of step (4) is as follows: For the underwater image containing a small number of pixel values reaching the limit value 0 or 255, the cumulative histogram statistics of the pixel values are performed before the affine transformation, the pixel points close to the limit value 0 or 255 in the histogram are removed, and the maximum value F of the remaining pixels is counted max and the minimum value F min ; the pixel values in the range [F min , F max ] are stretched to [0, 255] through affine transformation, and a clear image after underwater deblurring is obtained.

3. The underwater dark image processing method based on color space conversion according to claim 1, characterized in that, The implementation process of step (5) is as follows: By embedding the nonlinear channel prior into the traditional image deblurring model, the objective function is: Where a, b, and c are the corresponding weight parameters of the regularization term; For gradient operator; L is the potential sharp image; K is the blur kernel; C nl (L) represents the dark channel C of a potentially sharp image. d (L) and the bright channel C of the potentially sharp image b (L) ratio; J c The first constraint term uses the L2 norm to enhance the convolution result of the potentially sharp image L and the blur kernel K with the dehazed image J. c The first constraint term is the similarity of the two terms; the second constraint term is the correlation term of the nonlinear channel prior; the third constraint term is used to preserve the edge information of the image while smoothing out small details; the fourth term is the regularization term on K, which uses the traditional L2 norm to constrain the smoothness of the blur kernel.

4. The underwater low-light image processing method based on color space conversion according to claim 1, characterized in that, The light intensity of the water body was estimated using histogram statistical characteristics, as follows: The statistical average of the N pixels with the largest grayscale values ​​(0.05%) in the dark primary color image of the underwater image is taken as the water body light intensity value; the statistical average of the dark primary color values ​​of the original image is calculated as the water body illumination intensity value A. Where h is the horizontal axis value of the histogram; N is the total number of pixels in the dark primary color image.

5. The underwater low-light image processing method based on color space conversion according to claim 3, characterized in that, Applying constraints to K yields the following constraints: To improve the accuracy of the blur kernel calculation, K is estimated using the image gradient instead of the intensity: The optimal solution for K is obtained using the Fast Fourier Transform: Among them, C d (L) and C b (L) represent the dark and bright channels of the potentially sharp image, respectively; the initial restored image I is obtained using the obtained potentially sharp image L and the calculated precise blur kernel K. i .

Images