An underwater polarization image restoration method, program, device and storage medium for a suspended target in turbid water under artificial illumination of an AUV

Abstract

The present application belongs to the technical field of image processing, and particularly relates to an underwater polarization image restoration method for suspended targets in turbid water under artificial illumination for AUV, a program, an equipment and a storage medium. The present application realizes automatic identification of the background area through polarization characteristic difference, and then realizes automatic estimation of the backscattering polarization degree, eliminating artificial intervention; through multi-level collaborative estimation of the Gaussian pyramid, the original polarization characteristics of the image are avoided from being greatly damaged while noise is suppressed, and the backscattering estimation precision is improved; through frequency division processing of the Laplacian pyramid, the noise transmission path is further blocked in the model solving stage. The present application can more efficiently clarify the polarization image of the suspended target in the turbid water under the artificial illumination environment, has higher automation degree and complex environment adaptability, and is suitable for AUV online target detection.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically relating to a method, program, device, and storage medium for underwater polarization image restoration of suspended targets under artificial lighting in turbid waters by AUVs. Background Technology

[0002] Autonomous underwater vehicles (AUVs), as crucial platforms for independent deep-sea operations, rely heavily on their visual perception capabilities to determine the effectiveness of underwater target detection and identification. However, underwater light is affected by absorption by the water medium and scattering by suspended particles, leading to severe color distortion, fogging, reduced contrast, and noise in visual images, which worsens with increasing water turbidity. Furthermore, the absence of sunlight in deep-sea environments necessitates the use of artificial light. The imaging conditions and characteristics of artificial lighting differ from those under sunlight. Previous image reconstruction methods used under sunlight cannot be directly applied to artificial lighting. These factors significantly reduce the operational capabilities of AUVs in complex aquatic environments.

[0003] Underwater image sharpening methods are mainly divided into image enhancement and image restoration. Image enhancement does not rely on a physical imaging model and improves visual effects by directly adjusting pixel values, but its processing capability is limited in highly turbid waters and its environmental adaptability is poor. Image restoration, on the other hand, effectively removes scattering and improves contrast by accurately estimating key parameters and resolving the physical imaging model. Among these methods, underwater polarization imaging restoration technology separates the target information light from the backscattered light by analyzing the difference in polarization characteristics, enabling it to achieve imaging distances and material inversion capabilities. This is currently an important direction in the field of underwater image restoration. The Schechner model is the most widely accepted image restoration model in the field of underwater polarization restoration.

[0004] However, traditional restoration methods based on the Schechner model are mainly designed for clear water environments. When applied to the artificial lighting environment of turbid water bodies used in AUV operations, the following key problems exist:

[0005] First, it relies on manual interaction. Traditional methods require manual selection of a pure water background area to estimate the backscattered polarization degree, which cannot be automated. This prevents the algorithm from being embedded in the online real-time processing chain of AUVs, severely restricting its practical deployment on AUVs.

[0006] Second, the processing lacks noise resistance. The high concentration of suspended particles in turbid water causes the original image to contain a lot of noise. This noise is transmitted to the brightest and darkest images via the Stokes vector and is continuously amplified during the backscatter estimation process, resulting in inaccurate backscatter estimation and the inclusion of abnormal noise. At the same time, traditional methods do not perform noise reduction processing on the original image during the model solution stage, causing the noise to be further transmitted and amplified during the restoration process. The final restored image exhibits severe graininess, residual fog, and blurred details.

[0007] While the root cause of the aforementioned problems lies in noise, simply performing denoising on the original polarization image would disrupt the polarization relationships between pixels, leading to severe distortion of the polarization degree distribution and further reducing the restoration quality. Therefore, there is an urgent need for an automated polarization image restoration method that can suppress noise interference while avoiding significant damage to the original polarization characteristics. Summary of the Invention

[0008] The purpose of this invention is to solve the problems of artificial interaction dependence, inaccurate backscatter estimation, and noise propagation amplification in the existing technology when processing polarization images of turbid water bodies, and to provide a method, program, device and storage medium for underwater polarization image restoration of suspended targets under artificial lighting in turbid water bodies by AUV.

[0009] A method for underwater polarization image restoration of suspended targets under artificial lighting in turbid waters by AUVs includes the following steps:

[0010] Input the original image acquired by the polarization camera, calculate the Stokes vector of the original image, construct a reconstructed image at an arbitrary angle, take the reconstructed images with the highest and lowest average pixel brightness, and construct the polarization degree map of the original image.

[0011] Set the number of pyramid layers, construct a Gaussian pyramid from the original image polarization map, and calculate the backscattering polarization of each Gaussian pyramid image layer.

[0012] Gaussian pyramids are constructed for the reconstructed images with the highest and lowest average pixel brightness based on the pyramid layer number. The backscattering polarization degree of each layer of the Gaussian pyramid image is calculated based on the polarization degree map of the original image. The backscattering at each position in each layer of the Gaussian pyramid image is calculated. After upsampling, the backscattering of each pixel in the original image is obtained by weighted fusion.

[0013] A Laplacian pyramid is constructed from the original image based on the number of pyramid layers. For the highest-level Laplacian pyramid image, backscattering at each location in the original image is used for descattering and restoration. For Laplacian pyramid images of other layers, the image of that layer itself is used as a guide image for guided filtering.

[0014] The non-top-level Laplacian pyramid image after guided filtering is oversampled and superimposed with the top-level Laplacian pyramid image after descattering restoration to reconstruct the restored image; the restored image is then color-corrected to obtain a sharpened image.

[0015] Further, the calculation of the Stokes vector of the original image specifically involves:

[0016]

[0017] in, , and The Stokes vector of the original image; , , and Acquired by a polarization camera , , , Original polarization image;

[0018] Constructing a reconstructed image from any angle is specifically as follows:

[0019]

[0020] in, For angle The corresponding reconstructed image.

[0021] Furthermore, angle Corresponding reconstructed image Average pixel brightness for:

[0022]

[0023] in, and They are respectively direction and Number of pixels in the direction; , and angles Corresponding reconstructed image In the R, G and B channel images Pixel value at the location;

[0024] Reconstruct the image with the highest average pixel brightness Reconstructed image with the lowest average pixel brightness Construct the original image polarization map:

[0025]

[0026] in, In the polarization degree diagram of the original image Degree of polarization at the location; and The reconstructed image with the highest average pixel brightness Reconstructed image with the lowest average pixel brightness middle The pixel value at the location.

[0027] Furthermore, the setting of the number of pyramid layers A Gaussian pyramid is constructed from the polarization map of the original image, and the backscattering polarization degree of each layer of the Gaussian pyramid image is calculated, specifically as follows:

[0028] Take the first polarization degree map of the original image The G-channel image of the Gaussian pyramid is obtained by counting the number of pixels corresponding to each polarization degree in the G-channel image. The cumulative distribution curve of polarization degree in the G-channel image is obtained by accumulating the polarization degrees from smallest to largest and dividing by the total number of pixels. The polarization degree value corresponding to the steepest point on the cumulative distribution curve is taken as the polarization degree threshold. ;

[0029] Based on polarization degree threshold The pixel values ​​of pixels whose polarization degree exceeds the polarization degree threshold in the G channel image are set to 1, and the pixel values ​​of the remaining pixels in the G channel image are set to 0, thus obtaining a binary image of the background region mask.

[0030]

[0031] in, The first polarization map of the original image Background region mask in binary image of Gaussian pyramid image Pixel value at the location; The first polarization map of the original image G-channel image of a layered Gaussian pyramid Pixel value at the location;

[0032] The first polarization map of the background region mask binary image and the original image is compared. The Gaussian pyramid image is multiplied pixel by pixel to obtain the first polarization map of the original image. Background area of ​​the Gaussian pyramid image ;

[0033]

[0034] in, The first polarization map of the original image Background area of ​​the Gaussian pyramid image middle Pixel value at the location; The first polarization map of the original image In the image of the Gaussian pyramid Pixel value at the location;

[0035] Calculate the background area The average degree of polarization is obtained to get the first Backscattering polarization degree of layered Gaussian pyramid image ;

[0036]

[0037] in, The first polarization map of the original image Background area of ​​the Gaussian pyramid image Total number of pixels; The first polarization map of the original image In the image of the Gaussian pyramid The degree of polarization at the location.

[0038] Furthermore, the process of constructing Gaussian pyramids for the reconstructed images with the highest and lowest average pixel brightness based on the pyramid layer number, and calculating the backscattering at each location in each Gaussian pyramid image based on the backscattering polarization degree of each layer of the original image polarization degree map, specifically involves:

[0039]

[0040] in, For the first In the image of the Gaussian pyramid Backscattering at the location; The reconstructed image with the highest average pixel brightness The In the image of the Gaussian pyramid Pixel value at the location; The reconstructed image with the lowest average pixel brightness The In the image of the Gaussian pyramid Pixel value at the location;

[0041] Based on the backscattering at each location in the image of each layer of the Gaussian pyramid After upsampling and weighted fusion, the backscattered data of each pixel in the original image is obtained. ;

[0042]

[0043] in, Indicates upsampling; For the first The weights of backscattering in a layered Gaussian pyramid. .

[0044] Furthermore, for the image of the highest-level Laplacian pyramid, the descattering restoration is performed using backscattering at various locations in the original image, specifically as follows:

[0045]

[0046] in, Indicates downsampling; For the first part of the original image In the image of the layered Laplace pyramid Pixel value at the location; As the background light, the corresponding backscattered light from the original image is taken. The pixel value of the largest pixel;

[0047] For the Laplacian pyramid images that are not at the highest level, the image of that level itself is used as a guide image for guided filtering, specifically:

[0048] ,

[0049] in, For guided filtering processing; For the first part of the original image In the image of the layered Laplace pyramid The pixel value at the location.

[0050] Furthermore, the step of upsampling and superimposing the non-highest-level Laplacian pyramid image after guided filtering with the descattered and restored highest-level Laplacian pyramid image layer by layer to reconstruct the restored image specifically involves:

[0051] by As initial conditions, rebuild starting from the highest level:

[0052]

[0053] in, In the reconstructed image The pixel value at the location.

[0054] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method for underwater polarization image restoration of suspended targets under artificial illumination in turbid AUV waters.

[0055] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for underwater polarization image restoration of suspended targets under artificial lighting in turbid AUV waters.

[0056] A computer program product includes computer instructions that, when executed by a processor, implement the steps of the above-described method for underwater polarization image restoration of suspended targets under artificial illumination in turbid AUV waters.

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

[0058] This invention, based on the actual operational needs of AUVs, proposes a complete automated polarization image restoration technology solution. It achieves automatic background region identification through differences in polarization characteristics, thereby automatically estimating the backscattering polarization degree and eliminating manual intervention. Through multi-level Gaussian pyramid collaborative estimation, it suppresses noise while avoiding significant damage to the original polarization characteristics of the image, improving the accuracy of backscattering estimation. Furthermore, through Laplace pyramid frequency division processing, it further blocks noise propagation paths during the model solution stage. This invention can more efficiently and clearly restore the polarization images of suspended targets under artificial lighting conditions in turbid water bodies, exhibiting a higher degree of automation and adaptability to complex environments, making it suitable for online target detection by AUVs. Attached Figure Description

[0059] Figure 1 This is a diagram of an underwater polarization imaging model under artificial lighting in this invention.

[0060] Figure 2 This is the original polarization image of the suspended target (simulating a suspended anchor chain mine) used in the experiment in an embodiment of the present invention.

[0061] Figure 3 This is a comparison chart showing the image sharpening effects of the present invention and existing underwater image sharpening algorithms.

[0062] Figure 4 This table compares objective evaluation metrics among different algorithms.

[0063] Figure 5 A comparison table of time consumption for different algorithms (unit: seconds).

[0064] Figure 6 This is a diagram of the overall architecture of the present invention. Detailed Implementation

[0065] The present invention will now be further described with reference to the accompanying drawings.

[0066] This invention relates to the field of image processing, and is particularly applicable to a method for underwater polarization image restoration of suspended targets under artificial illumination in turbid water. This invention enables high-quality restoration of degraded suspended target polarization images acquired by AUVs under artificial illumination in turbid waters, meeting the real-time image processing requirements of AUV online target detection and achieving the clarification of degraded images.

[0067] A method for underwater polarization image restoration of suspended targets under artificial lighting in turbid waters by AUVs includes the following steps:

[0068] Step (1): Input the original image acquired by the polarization camera, calculate the Stokes vector and reconstruct it to obtain the brightest and darkest images, and then obtain the overall image polarization degree map based on the brightest and darkest images; then determine the threshold based on the cumulative distribution curve of the G channel polarization degree map, automatically identify the pure water background area through threshold segmentation, and calculate the mean polarization degree in the background area as the backscattering polarization degree.

[0069] This step consists of two parts: calculating the brightest and darkest images and the original polarization map, and automatically estimating the backscattered polarization. The detailed processing steps for each part are described below:

[0070] 1) Calculate the brightest and darkest images and the original polarization map.

[0071] Combination Figure 1 In turbid water environments under artificial lighting, the polarization characteristics of the target information light and backscattered light change after propagating through the water medium. The original image acquired by the polarization camera can be considered as a superposition of the target information light and backscattered light, and its polarization state can be fully described by the Stokes vector. Based on this imaging model, this step first calculates the Stokes vector of the original image acquired by the polarization camera. The calculation formula is as follows:

[0072]

[0073] In the formula: , and The Stokes vector of the original image; , , and These are four original polarization images at 0°, 45°, 90°, and 135°, acquired by a polarization camera.

[0074] Then the brightest image is calculated using the following formula. and the darkest image :

[0075]

[0076]

[0077] ;

[0078] In the formula: For any angle The corresponding reconstructed image; , and These are the R, G, and B channel images of the reconstructed image, respectively. For the corresponding polarization angle, ; The average pixel brightness of the image; The total number of pixels in the image; The brightest image; This is the darkest image.

[0079] get and Then, the original polarization map of the entire image can be calculated using the following formula:

[0080]

[0081] In the formula: Indicates the overall image polarization degree; The brightest image; This is the darkest image.

[0082] 2) Automatic estimation of backscattered polarization degree

[0083] This invention proposes an automatic estimation method for backscattered polarization degree based on differences in polarization characteristics. Experimental studies have revealed a significant difference in polarization degree between the target area and the background area under purely artificial light conditions, with the background area exhibiting a generally higher polarization degree and the target area exhibiting a generally lower polarization degree. Therefore, this difference is utilized to achieve automatic identification of the background area through threshold segmentation.

[0084] First, because underwater images are generally greenish, the G channel has the richest information, and the overall polarization map is smoother, making it easier to obtain an accurate polarization threshold. This invention chooses to use the G channel polarization map for background region identification. First, the G channel image of the overall image polarization map obtained in step 1) is extracted. The number of pixels corresponding to each polarization degree in the G channel polarization map is counted. Then, the polarization degree values ​​are accumulated from smallest to largest and divided by the total number of pixels to obtain the cumulative distribution curve of polarization degree. In the cumulative distribution curve, the polarization degree value corresponding to the steepest point of the curve is selected as the polarization threshold. Then, a background region mask is created according to the following formula:

[0085]

[0086] In the formula: Use it as a mask for the background area; This is a polarization diagram of the G channel; The polarization degree threshold is determined by the cumulative distribution curve of the polarization degree and is usually set to 0.7.

[0087] The background mask is a binary image, with pixel values ​​of 1 for the background region and 0 for the target region. Multiplying the mask pixel-by-pixel by the original image preserves the pixel values ​​of the background region while setting the pixel values ​​of the target region to zero, thus achieving the separation and extraction of the background region in pure water. The background region of the image can be obtained by multiplying the background mask by the original image, as shown in the following formula:

[0088]

[0089] In the formula: This refers to the background area of ​​the original image. Original image; Used as a mask for the background area.

[0090] After extracting the background region, the average polarization degree of each channel within the background region is calculated, which can then be used as an estimate of the polarization degree of the backscattered light. The calculation formula is shown below:

[0091]

[0092] In the formula: This is an estimate of the backscattering polarization degree; Indicates background area The total number of pixels in the image; Indicates the overall image polarization degree; This represents the background area of ​​the original image.

[0093] Step (2): Construct Gaussian pyramids with the same number of layers for the brightest image, darkest image and polarization degree map respectively. In each layer, the backscattering polarization degree is automatically estimated using the method in step (1). Then, backscattering is estimated using the brightest image, darkest image and backscattering polarization degree of that layer respectively. The backscattering of each layer is then upsampled to restore the original resolution and then fused according to the weight to obtain the fused backscattering.

[0094] This invention reveals that in clear water environments, image noise levels are low, and previous methods can effectively estimate backscatter and achieve good restoration results. However, when dealing with images of turbid water, image noise levels are high, and the method's backscatter estimation has two problems: first, the method is affected by noise interference, resulting in a shallower estimated backscatter, leading to residual fog in the final restored image; second, noise from the original image is transmitted and amplified into the backscatter, thus causing noise amplification in the restored image.

[0095] To address the aforementioned problems, this invention proposes a multi-level backscattering collaborative estimation method based on a Gaussian pyramid. The core idea is to construct a Gaussian pyramid of the same number of layers from the brightest image, darkest image, and polarization map. Backscattering polarization is estimated at different resolution levels to further estimate backscattering. Finally, a weighted fusion strategy is used to integrate the results from each level. This method leverages the noise attenuation characteristics of higher levels of the Gaussian pyramid through continuous filtering and downsampling, blocking the propagation path of noise from the original image to backscattering. Simultaneously, it utilizes the low-frequency characteristics of backscattering polarization to accurately extract global backscattering information at higher levels while preserving local structural details at lower levels. This achieves collaborative processing of "coarse-scale noise reduction and fine-scale fidelity preservation," obtaining a more accurate backscattering estimate without significantly damaging the original polarization characteristics of the image.

[0096] 1) Constructing the Gaussian Pyramid

[0097] The brightest image obtained in step (1) is obtained respectively. Darkest image And the original polarization diagram Construct Gaussian pyramids with the same number of layers.

[0098] In one embodiment of the present invention, the maximum number of layers is set to 2. After layer-by-layer Gaussian filtering and downsampling, layer 0 represents the original resolution, while the resolution of layers 1 and 2 is halved sequentially. The noise level in each layer also decreases layer by layer, while the backscattered polarization degree, as low-frequency information, is accurately extracted in higher layers.

[0099] 2) Calculate the backscattering polarization degree of each layer.

[0100] In polarization diagram Within each layer of the Gaussian pyramid, the backscattering polarization degree within each layer is independently calculated using the automatic estimation method in step (1), and the mean polarization degree is calculated within the background region corresponding to that layer.

[0101] Take the first polarization degree map of the original image The G-channel image of the Gaussian pyramid is obtained by counting the number of pixels corresponding to each polarization degree in the G-channel image. The cumulative distribution curve of polarization degree in the G-channel image is obtained by accumulating the polarization degrees from smallest to largest and dividing by the total number of pixels. The polarization degree value corresponding to the steepest point on the cumulative distribution curve is taken as the polarization degree threshold. ;

[0102] Based on polarization degree threshold The pixel values ​​of pixels whose polarization degree exceeds the polarization degree threshold in the G channel image are set to 1, and the pixel values ​​of the remaining pixels in the G channel image are set to 0, thus obtaining a binary image of the background region mask.

[0103]

[0104] in, The first polarization map of the original image Background region mask in binary image of Gaussian pyramid image Pixel value at the location; The first polarization map of the original image G-channel image of a layered Gaussian pyramid Pixel value at the location;

[0105] The first polarization map of the background region mask binary image and the original image is compared. The Gaussian pyramid image is multiplied pixel by pixel to obtain the first polarization map of the original image. Background area of ​​the Gaussian pyramid image ;

[0106]

[0107] in, The first polarization map of the original image Background area of ​​the Gaussian pyramid image middle Pixel value at the location; The first polarization map of the original image In the image of the Gaussian pyramid Pixel value at the location;

[0108] Calculate the background area The average degree of polarization is obtained to get the first Backscattering polarization degree of layered Gaussian pyramid image ;

[0109]

[0110] in, The first polarization map of the original image Background area of ​​the Gaussian pyramid image Total number of pixels; The first polarization map of the original image In the image of the Gaussian pyramid The degree of polarization at the location.

[0111] 3) Estimate backscattering within each level

[0112] Substituting the brightest and darkest images corresponding to each layer, along with the backscattering polarization degree of that layer, into the following formula, we can estimate the backscattering of each layer:

[0113]

[0114] in, For the first In the image of the Gaussian pyramid Backscattering at the location; The reconstructed image with the highest average pixel brightness The In the image of the Gaussian pyramid Pixel value at the location; The reconstructed image with the lowest average pixel brightness The In the image of the Gaussian pyramid Pixel value at the location;

[0115] 4) Weighted fusion of backscattering from each layer

[0116] To integrate the advantages of different levels, this invention employs a weighted fusion strategy to combine backscattering estimation results from multiple levels. After upsampling the higher-level results to restore them to their original resolution, they are fused with the lower-level results according to their weights to perform multi-level collaborative backscattering estimation. This achieves more accurate backscattering estimation while suppressing noise, thereby improving descattering performance and the quality of the restored image. The specific implementation formula is as follows:

[0117]

[0118] In the formula: Indicates upsampling; The weights for weighted fusion of backscattering at each level, and satisfying... .

[0119] In one embodiment of the present invention, the weights of layers 0, 1, and 2 are set to 0.15, 0.25, and 0.6, respectively, satisfying that the sum of the weights is 1. This allocation is based on the following criteria: layer 2 has the least noise and the most reliable global estimation, thus occupying the majority of the weight values; layer 0 has the most noise and is only used to supplement local details, therefore it has the lowest weight; layer 1 serves as a transition layer. The fused backscattering suppresses noise interference while preserving necessary local structure, effectively improving the problems of shallow scattering and noise residue in traditional methods.

[0120] Step (3): Construct a Laplacian pyramid for the original image, perform descattering restoration on the highest-level Laplacian pyramid image using the backscattering estimation results from step (2), perform guided filtering on the non-highest-level Laplacian pyramid images to suppress noise, and then upsample and stack the processed Laplacian pyramid images layer by layer to reconstruct the descattered restored image; finally, perform color correction on the reconstructed image to obtain the final sharpened image.

[0121] This invention has found that noise in backscattering has been effectively suppressed through multi-level backscattering co-estimation in step (2). However, if the unprocessed original intensity image and the estimated backscattering are directly substituted into the model for inverse solution, the noise in the original image will still be transmitted and amplified during the restoration process, resulting in a large number of abnormal noise and graininess in the final restored image.

[0122] To address the aforementioned problems, this invention proposes a descattering restoration method based on the Laplacian pyramid. The core idea is as follows: A Laplacian pyramid is constructed from the original intensity image. Utilizing its frequency band separation capability, the image is decomposed into a non-highest layer containing high-frequency noise and image edge details, and a highest layer containing low-frequency backscattering information. Guided filtering is applied to the non-highest layer to suppress noise and preserve edges, while descattering restoration is performed on the highest layer. Finally, a noise-free, dehazed restored image is obtained through pyramid reconstruction. The frequency band separation capability of the Laplacian pyramid enables frequency-division processing for noise reduction and restoration, improving the image restoration quality.

[0123] 1) Perform Laplacian pyramid decomposition on the original image.

[0124] To construct the Laplacian pyramid image of the original image, the highest number of layers must be consistent with the Gaussian pyramid parameters in step (2), and the highest number of layers is set to 2 as previously mentioned. Among them, layers 0 and 1 are non-highest layers, mainly containing high-frequency noise and image edge detail information; layer 2 is the highest layer, mainly containing low-frequency backscattering and information on the overall structure of the image.

[0125] 2) Guided filtering processing for non-top layers (layer 0, layer 1)

[0126] For the Laplacian pyramid images at layers 0 and 1 (i.e., non-highest layers), guided filtering is performed using the image of that layer itself as the guide image. Guided filtering can achieve smooth noise reduction under the constraints of a local linear model while maintaining sharp image edges. Its implementation formula is as follows:

[0127]

[0128] In the formula: For the first part of the original image Layered Laplacian pyramid image, in which That is, excluding the image of the highest level, the Laplace pyramid; For guided filtering, this expression represents the... Guided filtering is performed using itself as the guide image; This is the result of the guided filtering process.

[0129] 3) Top layer (2 layers) descattering restoration

[0130] Combination Figure 1 In the underwater polarization imaging model shown, backscattering is the main degradation factor causing the image fogging effect. In the highest layer (layer 2) of the Laplacian pyramid, low-frequency backscattering information dominates. Therefore, the fused backscattering estimated in step (2) can be directly used to descatter and restore the image of this layer according to the following formula to remove the fogging caused by backscattering:

[0131]

[0132] In the formula: The image is the highest layer of the original image, the Laplacian pyramid. The result of descattering restoration; The backscattering estimation result of step (2); and These represent upsampling and downsampling, respectively. For the background light, backscattering is selected. The pixel value of the brightest pixel is used as the background light estimate, i.e. .

[0133] 4) Pyramid Reconstruction

[0134] The non-highest-level Laplacian pyramid image after guided filtering is superimposed with the highest-level Laplacian pyramid image after descattering restoration, following the Laplacian pyramid reconstruction principle, and the reconstructed image is obtained by upsampling layer by layer. The reconstruction process starts from the highest level, upsampling layer by layer and adding to the corresponding level, finally restoring to the original resolution, resulting in the reconstructed image. The implementation formula is as follows:

[0135]

[0136] In the formula: To reconstruct the image for layer l, ,when At that time, it is the reconstructed image at the original resolution. ; Indicates upsampling; This is the processed image of the l-th layer of the Laplacian pyramid in this step (the images of non-top layers are the result after guided filtering, and the top layer is the result after descattering restoration). Pyramid reconstruction starts from the top layer, with the following initial conditions. .

[0137] 5) Color Correction

[0138] The reconstructed image is then processed using a grayscale world algorithm for color correction to correct color cast and obtain the final sharpened image.

[0139] Example 1:

[0140] To verify the effectiveness of the underwater polarization image restoration method for suspended targets under artificial lighting in turbid AUV waters designed in this patent, the algorithm proposed in this patent was experimentally compared with previous typical algorithms such as the Schechner algorithm by Schechner et al., the Treibitz algorithm by Treibitz et al., the PDCP algorithm by Boffety et al., the UDCP algorithm by Drews et al., the MIP algorithm by Bianco et al., and the IBLA algorithm by Peng et al. The original degraded underwater polarization images used in the experiments are attached. Figure 2 As shown in the attached figure, the underwater image sharpening effect is compared. Figure 3 As shown.

[0141] Figure 3 In the image, (a) corresponds to the unprocessed raw underwater RGB image; (b) corresponds to the image processed using the Schechner algorithm by Schechner et al. Figure 2 The result of the processing; (c) corresponding to the Treibitz algorithm by Treibitz et al. Figure 2 The result of the processing; (d) corresponding to the PDCP algorithm of Boffety et al. Figure 2 The result of the processing; (e) corresponding to the UDCP algorithm of Drews et al. Figure 2 The result of the processing; (f) corresponding to the MIP algorithm of Bianco et al. Figure 2 The result of the processing; (g) corresponding to the IBLA algorithm of Peng et al. Figure 2 The result of the processing; (h) corresponds to the algorithm of this invention. Figure 2 The result of the processing.

[0142] To more objectively evaluate the image quality of the algorithm's experimental results, this embodiment selects two commonly used underwater image quality evaluation metrics, UCIQE (Underwater Color Image Quality Evaluation) and NIQE (Natural Image Quality Evaluator), as the quality evaluation metrics for underwater image restoration. UCIQE reflects the linear quantitative evaluation result between color cast, blur, and contrast after underwater image restoration; a higher value indicates a better image processing effect. NIQE reflects the degree to which the statistical characteristics of the image closely resemble high-quality natural images; a lower value indicates a better image processing effect.

[0143] To demonstrate the time consumption of this invention compared to the above classic methods and algorithms, Figure 5 The algorithm execution time was compared statistically.

[0144] Quantitative analysis results of image restoration results from different algorithms, such as Figure 4 As shown, a comparative analysis of the evaluation metrics of various algorithms reveals that the algorithm proposed in this invention outperforms other underwater image sharpening algorithms in both the UCIQE and NIQE objective evaluation metrics. Therefore, for images of suspended targets under artificial lighting in turbid water, the algorithm proposed in this invention can better restore underwater images under artificial lighting compared to other underwater image sharpening algorithms.

[0145] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for underwater polarization image restoration of suspended targets under artificial illumination in turbid waters by AUVs, characterized in that: Input the original image acquired by the polarization camera, calculate the Stokes vector of the original image, construct a reconstructed image at an arbitrary angle, take the reconstructed images with the highest and lowest average pixel brightness, and construct the polarization degree map of the original image. Set the number of pyramid layers, construct a Gaussian pyramid from the original image polarization map, and calculate the backscattering polarization of each Gaussian pyramid image layer. Gaussian pyramids are constructed for the reconstructed images with the highest and lowest average pixel brightness based on the pyramid layer number. The backscattering polarization degree of each layer of the Gaussian pyramid image is calculated based on the polarization degree map of the original image. The backscattering at each position in each layer of the Gaussian pyramid image is calculated. After upsampling, the backscattering of each pixel in the original image is obtained by weighted fusion. The Laplacian pyramid is constructed from the original image based on the number of pyramid layers; for the image of the highest-level Laplacian pyramid, backscattering at each location in the original image is used for descattering and restoration. For images of Laplacian pyramids that are not at the highest level, the image of that level itself is used as a guide image for guided filtering. The non-top-level Laplacian pyramid image after guided filtering is oversampled and superimposed with the top-level Laplacian pyramid image after descattering restoration to reconstruct the restored image; the restored image is then color-corrected to obtain a sharpened image.

2. The method for underwater polarization image restoration of suspended targets under artificial illumination in turbid waters by AUVs, as described in claim 1, is characterized in that: The calculation of the Stokes vector of the original image is specifically as follows: in, , and The Stokes vector of the original image; , , and Acquired by a polarization camera , , , Original polarization image; Constructing a reconstructed image from any angle is specifically as follows: in, For angle The corresponding reconstructed image.

3. The method for underwater polarization image restoration of suspended targets under artificial illumination in turbid waters by AUVs, as described in claim 2, is characterized in that: angle Corresponding reconstructed image Average pixel brightness for: in, and They are respectively direction and Number of pixels in the direction; , and angles Corresponding reconstructed image In the R, G and B channel images Pixel value at the location; Reconstruct the image with the highest average pixel brightness Reconstructed image with the lowest average pixel brightness Construct the original image polarization map: in, In the polarization degree diagram of the original image Degree of polarization at the location; and The reconstructed image with the highest average pixel brightness Reconstructed image with the lowest average pixel brightness middle The pixel value at the location.

4. The method for underwater polarization image restoration of suspended targets under artificial illumination in turbid waters by AUVs, as described in claim 1, is characterized in that: Setting the number of pyramid layers A Gaussian pyramid is constructed from the polarization map of the original image, and the backscattering polarization degree of each layer of the Gaussian pyramid image is calculated, specifically as follows: Take the first polarization degree map of the original image The G-channel image of the Gaussian pyramid is obtained by counting the number of pixels corresponding to each polarization degree in the G-channel image. The cumulative distribution curve of polarization degree in the G-channel image is obtained by accumulating the polarization degrees from smallest to largest and dividing by the total number of pixels. The polarization degree value corresponding to the steepest point on the cumulative distribution curve is taken as the polarization degree threshold. ; Based on polarization degree threshold The pixel values ​​of pixels whose polarization degree exceeds the polarization degree threshold in the G channel image are set to 1, and the pixel values ​​of the remaining pixels in the G channel image are set to 0, thus obtaining a binary image of the background region mask. in, The first polarization map of the original image Background region mask in binary image of Gaussian pyramid image Pixel value at the location; The first polarization map of the original image G-channel image of a layered Gaussian pyramid Pixel value at the location; The first polarization map of the background region mask binary image and the original image is compared. The Gaussian pyramid image is multiplied pixel by pixel to obtain the first polarization map of the original image. Background area of ​​the Gaussian pyramid image ; in, The first polarization map of the original image Background area of ​​the Gaussian pyramid image middle Pixel value at the location; The first polarization map of the original image In the image of the Gaussian pyramid Pixel value at the location; Calculate the background area The average degree of polarization is obtained to get the first Backscattering polarization degree of layered Gaussian pyramid image ; in, The first polarization map of the original image Background area of ​​the Gaussian pyramid image Total number of pixels; The first polarization map of the original image In the image of the Gaussian pyramid The degree of polarization at the location.

5. The underwater polarization image restoration method for suspended targets under artificial illumination in turbid waters by AUVs, as described in claim 4, is characterized in that: The process involves constructing Gaussian pyramids for the reconstructed images with the highest and lowest average pixel brightness based on the pyramid layer number, and calculating the backscattering polarization at each location in each Gaussian pyramid image based on the backscattering polarization degree of each layer of the original image polarization degree map. Specifically: in, For the first In the image of the Gaussian pyramid Backscattering at the location; The reconstructed image with the highest average pixel brightness The In the image of the Gaussian pyramid Pixel value at the location; The reconstructed image with the lowest average pixel brightness The In the image of the Gaussian pyramid Pixel value at the location; Based on the backscattering at each location in the image of each layer of the Gaussian pyramid After upsampling and weighted fusion, the backscattered data of each pixel in the original image is obtained. ; in, Indicates upsampling; For the first The weights of backscattering in a layered Gaussian pyramid. .

6. The underwater polarization image restoration method for suspended targets under artificial illumination in turbid waters by AUVs, as described in claim 5, is characterized in that: For the image of the highest-level Laplacian pyramid, the backscattering at various locations in the original image is used for descattering and restoration, specifically as follows: in, Indicates downsampling; For the first part of the original image In the image of the layered Laplace pyramid Pixel value at the location; As the background light, the corresponding backscattered light from the original image is taken. The pixel value of the largest pixel; For the Laplacian pyramid images that are not at the highest level, the image of that level itself is used as a guide image for guided filtering, specifically: ， in, For guided filtering processing; For the first part of the original image In the image of the layered Laplace pyramid The pixel value at the location.

7. The underwater polarization image restoration method for suspended targets under artificial illumination in turbid waters by AUVs, as described in claim 6, is characterized in that: The process of upsampling and superimposing the non-highest-level Laplacian pyramid image after guided filtering with the descattered and restored highest-level Laplacian pyramid image layer by layer to reconstruct the restored image is as follows: by As initial conditions, rebuild starting from the highest level: in, In the reconstructed image The pixel value at the location.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7.

10. A computer program product comprising computer instructions, characterized in that: When executed by a processor, the computer instructions implement the steps of the method according to any one of claims 1 to 7.

Images