Deep seabed weak illumination image enhancement method based on non-uniform illumination correction

Abstract

The invention discloses a deep seabed weak illumination image enhancement method based on non-uniform illumination correction. The deep seabed weak illumination image enhancement method comprises the following steps: designing a non-linear guide filtering algorithm to realize balance of a bright region and a dark region of an underwater image; an adaptive MSRCR algorithm is adopted to eliminate artifacts of the underwater image; and carrying out multi-scale image fusion on the basis of the Retinex model to obtain an enhanced clear image. The method has good adaptability in the underwater environment with non-uniform illumination and low illumination, the influence of non-uniform illumination on the image can be corrected, the brightness and color of the underwater image can be effectively recovered, artifacts of the image can be adaptively eliminated, and technical support is provided for high-quality visual perception of a deep sea exploration sampling vehicle.

Description

Technical Field

[0001] This invention relates to the field of visual perception technology, and in particular to a method for enhancing deep-sea images in low-light conditions based on non-uniform illumination correction. 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. Visual images are an important way for exploration and sampling vehicles to observe the surrounding environment of the deep seabed. Due to their advantages such as high information density and fast perception speed, they are widely used in deep-sea exploration and mining equipment. 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 three-dimensional topographic visual reconstruction. However, the deep-sea environment is complex and often has various influencing factors. Artificial light sources can increase underwater visibility distance, but under direct illumination, the uneven distribution of light causes bright spots in the underwater images, resulting in inconsistent overall image brightness and seriously affecting the visual effect of the exploration and sampling vehicle.

[0003] Currently, many scholars have conducted research on underwater image enhancement methods in low-light conditions, which can be broadly categorized into the following three types:

[0004] The first method is histogram equalization (HE), which evenly distributes the pixels of an image, expanding the dynamic range of the pixels and thus improving the contrast of low-light images. Based on this principle, many researchers have extended its application. While these methods improve the contrast of low-light images, they do not consider the actual lighting conditions and introduce more noise in the process of improving contrast.

[0005] The second type is low-light image enhancement methods based on image fusion. Common methods include nonlinear transformation methods based on pixel-level operations. These methods often rely on nonlinear transformation functions, such as gamma correction, sigmoid transfer function, and logarithmic transfer function. These image enhancement methods primarily perform pixel-level operations on low-light images and have achieved some success. However, the parameters of the nonlinear transformation function are difficult to determine, resulting in inconsistent enhancement effects. Furthermore, fusion-based methods are often limited in practical applications due to the lack of a fusion source.

[0006] The third approach is low-light image enhancement based on the Retinex model. Retinex theory is a color perception model based on human vision, exhibiting a degree of color constancy. Retinex-based algorithms have shown better performance in enhancing low-light images compared to other methods, including the classic Single-Scale Retinex (SSR) and Multi-Scale Retinex (MSR) algorithms. Classical Retinex algorithms often consider using a center / surround function for illumination estimation, but this approach struggles to strike a good balance between dynamic range compression and color constancy, and also introduces halo artifacts into the enhanced image. To address these issues, many researchers have improved upon Retinex theory, primarily by refining the illumination and reflectance maps of the Retinex model. However, these methods enhance low-light images by eliminating illumination, leading to image information loss, and inefficient image decomposition introduces various dark region artifacts into the enhanced image. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a simple and adaptable method for enhancing deep-sea images in low-light conditions based on non-uniform illumination correction.

[0008] The technical solution of this invention to solve the above-mentioned technical problems is: a method for enhancing deep-sea images in low-light conditions based on non-uniform illumination correction, comprising the following steps:

[0009] Step 1: Design a nonlinear guided filtering algorithm to achieve equalization of bright and dark areas in underwater images;

[0010] Step 2: Use the adaptive MSRCR algorithm to eliminate artifacts in underwater images;

[0011] Step 3: Perform multi-scale image fusion based on the Retinex model to obtain an enhanced and clear image.

[0012] The above-mentioned deep-sea low-light image enhancement method based on non-uniform illumination correction, the specific process of step 1 is as follows:

[0013] Step 11: Convert the original image from RGB color space to HSV color space. The expression is:

[0014]

[0015] In the formula, , , These are the red, green, and blue channels in the RGB color space, respectively. , , These are the hue channel, saturation channel, and brightness channel of the HSV color space, respectively. The function is for finding the minimum value; This is a function to find the maximum value.

[0016] Step 12: Use guided filtering GIFs Perform the operation to obtain the initial illumination component;

[0017] Step 13: Design a nonlinear guided filter illumination equalization algorithm to achieve equalization of bright and dark areas;

[0018] Step 14: Introduce a dark channel prior algorithm to process the initial illumination component. By obtaining the dark channel weight map of the initial illumination component, the influence of non-uniform illumination on the underwater image is further eliminated, resulting in a sub-image. ; divide the image The image is segmented into blocks, and the cumulative distribution function of each region is calculated. Sharp grayscale histograms are then cut, and the number of pixels from the cut-out blocks is redistributed evenly across the grayscale histogram to obtain the final image. .

[0019] The specific process of step 13 in the above-mentioned deep-sea low-light image enhancement method based on non-uniform illumination correction is as follows:

[0020] Adaptive brightness correction is performed using the Weber-Fechner law, which states that there is a logarithmic linear relationship between subjective and objective brightness perception in the human eye.

[0021]

[0022] In the formula, This is the component image after nonlinear guided filtering. For objective brightness perception, and All are constants. This is a logarithmic transformation relationship;

[0023] By fitting the above equation with a function, we obtain:

[0024]

[0025] In the formula, To adjust the coefficient, , For saturation component image The average value, This is the image after guided filtering.

[0026] The above-mentioned deep-sea low-light image enhancement method based on non-uniform illumination correction, the specific process of step 2 is as follows:

[0027] Step 21: Import color restoration parameters Adjust the aspect ratio of the three channels of the underwater image to obtain the adjusted image. ;

[0028] Step 22: Adjust the image Transform to HSV space, by... The operation dynamically adjusts the probability density of its pixels to improve image detail blur and artifacts. It calculates the probability density of each intensity level in the underwater image to adaptively represent the intensity changes in the image, thus obtaining the reflection component. ;

[0029] Step 23: Use the grayscale world white balance algorithm to adjust the reflection component. The underwater image is processed to restore its true colors, resulting in the image. .

[0030] In the above-mentioned deep-sea low-light image enhancement method based on non-uniform illumination correction, step 21 introduces a color restoration parameter. The formula for adjusting the three-channel scale of an underwater image is as follows:

[0031]

[0032]

[0033] In the formula, For network multi-scale retinal enhancement functions, Input image; For the image number Color recovery parameters for each channel; For the number of scales; Index for the number of scales; For the first Weighting coefficients for each scale; It is a Gaussian filter function. For the input image, the first One channel component; and These are all empirical parameters; This represents the number of channels in the image.

[0034] In the above-described deep-sea low-light image enhancement method based on non-uniform illumination correction, the formula for calculating the probability density of each intensity level of the underwater image in step 22 is as follows:

[0035]

[0036] In the formula, It is a weighted probability density distribution; Image intensity The weighted probability density distribution; To adjust the parameters; Image intensity The probability density; It is the probability density with the highest value; It is the probability density with minimum value.

[0037] In the above-mentioned deep-sea low-light image enhancement method based on non-uniform illumination correction, step 23 involves estimating the color distribution of the light source by calculating the average grayscale value of each color channel, expressed as:

[0038]

[0039]

[0040] In the formula, The weight of the red channel; Weighting for the green channel; The weight of the blue channel; This represents the average value of the red channel. This represents the average value of the green channel. This represents the average value of the blue channel. , , They are respectively , , Channel values; , and After adjustment , , grayscale value, This represents the average grayscale value of the RGB three-channel components.

[0041] The above-mentioned deep-sea low-light image enhancement method based on non-uniform illumination correction, specifically step 3, is as follows:

[0042] Step 31: Put , and As the first input for multi-scale image fusion, the corrected illumination component is obtained after processing.

[0043] Step 32: , and As the second input for multi-scale image fusion, the corrected reflection component is obtained after processing.

[0044] Step 33: Using the corrected illumination and reflection components, the enhanced underwater clear image is obtained based on the Retinex model.

[0045] The above-mentioned deep-sea weak-light image enhancement method based on non-uniform illumination correction, in step 31, the processing procedure is as follows: calculate the brightness weight, color weight, and average weight of the first input respectively, then calculate the Laplacian value of the first input and the Gaussian value of the corresponding weight mapping, and finally perform multi-scale operation to obtain the corrected illumination components.

[0046] In the above-mentioned deep-sea weak-light image enhancement method based on non-uniform illumination correction, step 32 involves the following steps: calculating the Laplacian contrast weight, dark channel weight, exposure weight, and saturation weight of the second input respectively; then obtaining the Laplacian value of the second input and the Gaussian value mapped by the corresponding weights; and finally obtaining the corrected reflection component through multi-scale fusion.

[0047] The beneficial effects of this invention are as follows:

[0048] 1. This invention has good adaptability in underwater environments with non-uniform lighting and low light, can correct the impact of non-uniform lighting on images, and can also effectively restore the brightness and color of underwater images and adaptively eliminate image artifacts.

[0049] 2. This invention designs a nonlinear guided filtering algorithm to achieve equalization of bright and dark areas in underwater images, and uses an adaptive MSRCR algorithm to eliminate image artifacts. Experimental results show that this invention can adapt well to non-uniform low-light underwater environments, providing technical support for high-quality visual perception of deep-sea exploration and sampling vehicles. Attached Figure Description

[0050] Figure 1 This is the overall flowchart of the present invention.

[0051] Figure 2 This is a schematic diagram of the original image.

[0052] Figure 3 This is a schematic diagram of the luminance channel.

[0053] Figure 4 This is a schematic diagram of guided filtering applied to the original image.

[0054] Figure 5 This is a schematic diagram of applying nonlinear guided filtering to the original image.

[0055] Figure 6 This is an enlarged view of the portion captured from the original image.

[0056] Figure 7 This is a schematic diagram of the adjusted image.

[0057] Figure 8 This is a schematic diagram of the reflection component.

[0058] Figure 9 This is a schematic diagram of an image converted from the reflection component to RGB space.

[0059] Figure 10 This is a schematic diagram of the process of estimating illumination components in multi-scale fusion, where (a) is the non-uniform illumination correction image, (b) is the brightness weight map, (c) is the color weight map, (d) is the fusion weight map, and (e) is the regularization weight map.

[0060] Figure 11 This is a schematic diagram of the reflection component estimation process in multi-scale fusion, where (f) is the image after artifact adaptive removal, (g) is the Laplacian weight map, (h) is the dark channel weight map, (i) is the exposure weight map, and (j) is the saturation weight map.

[0061] Figure 12 This is a schematic diagram of the corrected illumination components.

[0062] Figure 13 This is a schematic diagram of the corrected reflection components.

[0063] Figure 14 This is a schematic diagram of an image enhanced using the present invention. Detailed Implementation

[0064] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0065] like Figure 1 As shown, the deep-sea low-light image enhancement method based on non-uniform illumination correction includes the following steps:

[0066] Step 1: Design a nonlinear guided filtering algorithm to achieve equalization of bright and dark areas in underwater images. The specific process is as follows:

[0067] Step 11: Convert the original image from RGB color space to HSV color space. The expression is:

[0068]

[0069] In the formula, , , These are the red, green, and blue channels in the RGB color space, respectively. , , These are the hue channel, saturation channel, and brightness channel of the HSV color space, respectively. The function is for finding the minimum value; This is a function to find the maximum value. The original image is as follows: Figure 2 As shown, the luminance channel of the HSV color space like Figure 3 As shown.

[0070] Step 12: Use guided filtering GIFs Perform operations to obtain the initial illumination components, such as Figure 4 As shown.

[0071] Step 13: Design a nonlinear guided filter illumination equalization algorithm to achieve equalization between bright and dark areas, such as... Figure 5 As shown.

[0072] Adaptive brightness correction is performed using the Weber-Fechner law, which states that there is a logarithmic linear relationship between subjective and objective brightness perception in the human eye.

[0073]

[0074] In the formula, This is the component image after nonlinear guided filtering. For objective brightness perception, and All are constants. This is a logarithmic transformation relationship;

[0075] To avoid increasing computational costs due to logarithmic operations, and also to prevent excessive image enhancement, a function is used to fit the above equation, resulting in:

[0076]

[0077] In the formula, To adjust the coefficient, The smaller the value, the larger the adjustment range, which can be determined based on the saturation component image. Calculate the average value The value, , For saturation component image The average value, This is the image after guided filtering.

[0078] Step 14: Introduce a dark channel prior algorithm to process the initial illumination component. By obtaining the dark channel weight map of the initial illumination component, the influence of non-uniform illumination on the underwater image is further eliminated, resulting in a sub-image. ; divide the image The image is segmented into blocks, and the cumulative distribution function of each region is calculated. Sharp grayscale histograms are then cut, and the number of pixels from the cut-out blocks is redistributed evenly across the grayscale histogram to obtain the final image. .

[0079] Step 2: Use the adaptive MSRCR algorithm to eliminate artifacts in underwater images. The specific process is as follows:

[0080] Step 21: Import color restoration parameters Adjust the aspect ratio of the three channels of the underwater image to obtain the adjusted image. (like Figure 7 As shown in the figure, this effectively solves the problem of color distortion in the original image.

[0081] Introducing color restoration parameters The formula for adjusting the three-channel scale of an underwater image is as follows:

[0082]

[0083]

[0084] In the formula, For network multi-scale retinal enhancement functions, Input image; For the image number The color recovery parameters for each channel are mainly used to adjust the proportions of different color channels. For the number of scales, ; Index for the number of scales; For the first Weighting coefficients for each scale; It is a Gaussian filter function. For the input image, the first One channel component; and These are all empirical parameters. Take 46, Take 125; The number of channels in the image. Take 3.

[0085] Step 22: Adjust the image Transform to HSV space, by... The operation dynamically adjusts the probability density of its pixels to improve image detail blur and artifacts. It calculates the probability density of each intensity level in the underwater image to adaptively represent the intensity changes in the image, thus obtaining the reflection component. (like Figure 8 (as shown); reflection component Images converted to RGB space, such as Figure 9 As shown.

[0086] The formula for calculating the probability density of each intensity level in an underwater image is:

[0087]

[0088] In the formula, It is a weighted probability density distribution; Image intensity The weighted probability density distribution; To adjust the parameters, set it to 0.3, mainly for slightly modifying the statistical histogram and reducing the generation of adverse factors; Image intensity The probability density; It is the probability density with the highest value; It is the probability density with minimum value.

[0089] Step 23: Use the grayscale world white balance algorithm to adjust the reflection component. The underwater image is processed to restore its true colors, resulting in the image. .

[0090] The color distribution of the light source is estimated by calculating the average gray value of each color channel, as expressed by the following expression:

[0091]

[0092]

[0093] In the formula, The weight of the red channel; Weighting for the green channel; The weight of the blue channel; This represents the average value of the red channel. This represents the average value of the green channel. This represents the average value of the blue channel. , , They are respectively , , Channel values; , and After adjustment , , grayscale value, This represents the average grayscale value of the RGB three-channel components.

[0094] Step 3: Perform multi-scale image fusion based on the Retinex model to obtain an enhanced, clearer image. The specific process is as follows:

[0095] Step 31: Put , and As the first input for multi-scale image fusion, the brightness weight, color weight, and average weight of brightness and color are calculated for the first input, respectively. Figure 10 As shown, Figure 10 In the image, (a) is the non-uniform illumination correction image, (b) is the brightness weight map, (c) is the color weight map, (d) is the fusion weight map, and (e) is the regularization weight map. Then, the Laplacian value of the first input and the corresponding Gaussian values ​​of the weight mapping are calculated. Finally, multi-scale operations are performed to obtain the corrected illumination components, as shown in the image. Figure 12 As shown;

[0096] Step 32: , and As the second input for multi-scale image fusion, the Laplacian contrast weight, dark channel weight, exposure weight, and saturation weight of the second input are calculated respectively, such as... Figure 11 As shown, Figure 11 In the image, (f) is the image after adaptive artifact removal, (g) is the Laplacian weight map, (h) is the dark channel weight map, (i) is the exposure weight map, and (j) is the saturation weight map. Next, the Laplacian values ​​of the second input and the corresponding Gaussian values ​​of the weight mappings are obtained. Finally, multi-scale fusion is used to obtain the corrected reflection components, as shown in the image. Figure 13 As shown;

[0097] Step 33: Using the corrected illumination and reflection components, and based on the Retinex model, obtain the enhanced underwater clear image, such as... Figure 14 As shown.

[0098]

[0099]

[0100] Table 1 shows the quantitative comparison results of different image enhancement algorithms for low-light pool scenes. The bolded or underlined values ​​in the table are the optimal values ​​of the corresponding algorithms. The data in the table shows that the information entropy (IE) of this invention is 7.9128, the average gradient (AVG) is 3.9434, the underwater color image quality evaluation (UCIQE) is 0.6655, and the image naturalness (NIQE) is 2.7829. This invention outperforms the other six classic algorithms in IE, AVG, UCIQE, and NIQE. The six classic algorithms are UDCP (Underwater Dark Channel Prior), IBLA (Underwater Image Restoration Based on Image Blur and Light Absorption), WCID (Wavelength Compensation and Image Dehazing), BIMEF (Bio-Inspired Multiple Exposure Fusion), FWE (Fusion-Based Low-Light Image Enhancement), and SDD (Semi-Decoupled Decomposition Low-Light Image Enhancement). This indicates that the image enhanced by this invention has more natural color restoration, better fidelity, higher clarity, more information, and better visual effects.

[0101] Table 2 shows the qualitative comparison results of different methods for the low-light underwater image dataset. The information entropy (IE) of this invention is 7.8921, the average gradient (AVG) is 9.3219, the underwater color image quality evaluation (UCIQE) is 0.6037, and the image naturalness (NIQE) is 3.2313. The invention outperforms the other six classic algorithms in terms of IE, AVG, UCIQE, and NIQE, indicating that the visual effect of this invention is better.

Claims

1. A deep-sea weak-light image enhancement method based on non-uniform light correction, characterized in that, The method comprises the following steps: Step 1: design a nonlinear guided filter algorithm to realize the equalization of bright and dark regions of the underwater image; Step 2: adopt an adaptive MSRCR algorithm to eliminate the artifacts of the underwater image; Step 3: perform multi-scale image fusion on the basis of the Retinex model to obtain an enhanced clear image.

2. The method according to claim 1, wherein, The specific process of step 1 is as follows: Step 11: convert the original image from the RGB color space to the HSV color space, and the expression is as follows: ； wherein, , , R, G, B are the red, green, blue color channels of the RGB color space, respectively; , , H, S, V are the hue, saturation, value color channels of the HSV color space, respectively; is a minimum function; is a maximum function; Step 12: Perform operation on the guided filter GIF to obtain an initial illumination component; Step 13: design a light equalization algorithm of the nonlinear guided filter to realize the equalization of bright and dark regions; Step 14: introducing the dark channel prior algorithm to process the initial illumination component, further eliminating the influence of non-uniform illumination on the underwater image by calculating the dark channel weight map of the initial illumination component, to obtain a sub-image ; the sub-image is processed in blocks, the cumulative distribution function of each region is calculated, the sharp gray histogram is cut, and the number of cut pixels is uniformly redistributed on the gray histogram, and finally the image is obtained.

3. The method of claim 2, wherein the method is based on non-uniform illumination correction for deep-sea weak-light image enhancement. The specific process of step 13 is as follows: Adopt the Weber-Fechner law to perform adaptive brightness correction, and there is a logarithmic linear relationship between the subjective brightness perception and the objective brightness perception of the human eye, that is: ； wherein is a component map after a non-linear guided filtering process, is an objective luminance perception, and are constants, is a logarithmic transformation relationship; Adopt a function to fit the above formula, and the result is as follows: ； In the formula, is an adjustment coefficient, , is a saturated component image is an average value, is an image after the guided filter processing.

4. The method according to claim 3, wherein, The specific process of step 2 is as follows: Step 21: introducing color restoration parameters adjusting the three-channel proportion of the underwater image to obtain an adjusted image ; Step 22: Adjusted image Transforming to HSV space, by operating on the probability density of its pixels to improve the image's detail blur and artifacts, the probability density of each intensity level of the underwater image is calculated to adaptively represent the intensity variation of the image, resulting in a reflectance component ; Step 23: Apply gray world white balance algorithm to the reflectance component to recover the true colors of the underwater image, resulting in image .

5. The method of claim 4, wherein the method is based on non-uniform illumination correction for deep-sea weak light enhancement. In step 21, the color restoration parameter is introduced Adjust the three-channel proportion of the underwater image, which is expressed by the formula: ； ； wherein is a network multiscale retinal enhancement function, is an input image; is a color restoration parameter for the i-th channel of the image; is a number of scales; is an index of the number of scales; is a weighting coefficient for the i-th scale; is a Gaussian filter function, is an i-th channel component of the input image; and and are empirical parameters; is a number of channels of the image.

6. The method of claim 5, wherein the method is based on non-uniform illumination correction for deep-sea weak light enhancement. In step 22, the formula for calculating the probability density of each intensity level of the underwater image is as follows: ； In the formula, It is a weighted probability density distribution; Image intensity The weighted probability density distribution; To adjust the parameters; Image intensity The probability density; It is the probability density with the highest value; It is the probability density with minimum value.

7. The method according to claim 6, wherein, In step 23, the color distribution of the light source is estimated by calculating the average gray value of each color channel, and the expression is as follows: ； ； wherein, is the weight of the red channel; is the weight of the green channel; is the weight of the blue channel; is the average value of the red channel; is the average value of the green channel; is the average value of the blue channel; , , are the channel values of , , , and are the adjusted , , gray values of is the average gray value of the RGB three-channel components.​ 8. The method according to claim 7, wherein, The specific process of step 3 is as follows: Step 31: obtaining the modified illumination component by processing the first input of multiscale image fusion , and ​ Step 32: obtaining the modified reflection component by processing the second input of multiscale image fusion , and ​ Step 33: use the corrected light component and the reflection component, and then obtain the enhanced underwater clear image according to the Retinex model.

9. The method of claim 8, wherein the method is based on non-uniform illumination correction for deep-sea weak light enhancement. In step 31, the process is as follows: calculate the brightness weight, color weight, average weight of brightness and color of the first input respectively, then calculate the Laplacian value of the first input and the corresponding weight-mapped Gaussian value, and finally perform multi-scale operation to obtain the corrected light component.

10. The method of claim 8, wherein the method is based on non-uniform illumination correction for deep-sea weak light image enhancement. In step 32, the process is as follows: calculate the Laplacian contrast weight, dark channel weight, exposure weight and saturation weight of the second input respectively, then obtain the Laplacian value of the second input and the corresponding weight-mapped Gaussian value, and finally perform multi-scale fusion to obtain the corrected reflection component.

Images