Underwater image enhancement method and system based on polarization difference imaging

Abstract

The invention provides an underwater image enhancement method and system based on polarization difference imaging, and relates to the technical field of image processing, and the method comprises the steps: obtaining a depth image corresponding to a to-be-enhanced underwater image and two polarization images with the polarization degree exceeding a set threshold value, and carrying out the calculation to obtain a difference image between the two polarization images; based on the difference image, performing region division on the two polarization images, and calculating the background light intensity of each partition after division by using the depth image; calculating and optimizing the perspective rate of each pixel point in the corresponding partition based on the background light intensity; and recovering the image content of the corresponding partition of the corresponding polarization image based on the perspective rate, and carrying out weighted fusion and global image enhancement processing on the two recovered polarization images according to the polarization degree to obtain a final enhanced image. The background light intensity and the perspective rate of each partition can be estimated more accurately, so that the underwater image is effectively enhanced.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an underwater image enhancement method and system based on polarization differential imaging. Background Art

[0002] In the field of underwater imaging, obtaining clear, high-quality underwater images has always been a technical challenge due to the absorption and scattering of water and changes in lighting conditions. Traditional underwater image enhancement methods, such as histogram equalization and contrast stretching, can improve the visual effect of images to a certain extent, but they are often unable to effectively remove blur and color distortion caused by the underwater environment. In recent years, with the development of polarization imaging technology, methods that use polarization information to enhance underwater images have gradually become a research hotspot.

[0003] Polarization imaging technology can provide more information than traditional imaging by capturing the polarization state of light waves, especially in dealing with scattered light and background light interference. In underwater environments, due to the presence of scattered light, the target and background in the image are often difficult to distinguish. Polarization imaging technology can distinguish between scattered light and direct light by measuring the polarization state of light waves, thereby effectively reducing the impact of scattered light and improving image clarity.

[0004] However, most of the existing underwater image enhancement methods based on polarization imaging only use image information in a single polarization state, ignoring the differences between images in different polarization states, which limits the improvement of image enhancement effects. In addition, when dealing with complex underwater scenes, these methods often have difficulty accurately estimating the background light intensity and perspective, resulting in deviations in image restoration results.

[0005] Therefore, it is necessary to provide an underwater image enhancement method and system based on polarization differential imaging to solve the above technical problems. Summary of the invention

[0006] In order to solve the above technical problems, the present invention provides an underwater image enhancement method and system based on polarization differential imaging, which obtains a depth image corresponding to the underwater image to be enhanced and two polarization images with polarization degrees exceeding a set threshold, and uses the difference image between the two polarization images to perform region division and background light intensity calculation. By combining the information of the depth image and the difference image, the method can more accurately estimate the background light intensity and perspective of each partition, thereby achieving effective enhancement of the underwater image. The present invention provides an underwater image enhancement method based on polarization differential imaging, the method comprising the following steps: Acquire a depth image corresponding to the underwater image to be enhanced and two polarization images whose polarization degree exceeds a set threshold, and calculate a difference image between the two polarization images; Based on the difference image, the two polarization images are divided into regions respectively, and the background light intensity of each divided region is calculated using the depth image; Based on the background light intensity, calculating and optimizing the perspective rate of each pixel in the corresponding partition; The image content of the corresponding partition of the corresponding polarization image is restored based on the perspective, and the two restored polarization images are weightedly fused and globally enhanced according to the polarization degree to obtain a final enhanced image.

[0007] Preferably, the partitions include a target area, a background area and a transition area.

[0008] Preferably, the acquisition of the polarization image includes: Two polarization images are acquired in sequence at different polarization degrees using the same shooting device, and the polarization degree difference exceeds a set threshold.

[0009] Preferably, the acquisition of the difference image includes: For two polarization images, pixel-level matching is performed at the same spatial position; For each pair of matching pixels, the difference between the pixel values ​​of the two polarization images is calculated and the absolute value is taken to obtain a difference image.

[0010] Preferably, the dividing the two polarization images into regions based on the difference image comprises: Processing the difference image based on a deep learning semantic segmentation model to identify and mark the target area, background area and transition area; The marking results are mapped to the corresponding two polarization images, and the two polarization images are divided into a target area, a background area, and a transition area.

[0011] Preferably, the step of calculating the background light intensity of each divided partition using the depth image includes: For the background area, the pixel difference values ​​of the corresponding area in the difference image are extracted, the variance of the pixel difference values ​​is calculated and used as a weight factor, and the pixel values ​​of the depth image are weighted averaged to obtain the background light intensity of the background area; For the target area, the polarization attenuation coefficient is determined according to the polarization degree ratio of the two polarization images in the target area. Combined with the local gradient information of the depth image, the adaptive weight is generated through the normalized product function to calculate the background light intensity of the target area. For the transition area, edge detection is performed on the difference image and a threshold comparison is performed with the pixel intensity difference of the two polarization images to generate a mask. Based on the mask, bidirectional linear interpolation is performed on the background light intensity of the background area and the target area to obtain the background light intensity of the transition area.

[0012] Preferably, the calculating and optimizing the perspective rate of each pixel in the corresponding partition based on the background light intensity includes: For each pixel in each partition, the preliminary perspective is calculated based on the background light intensity and the original brightness value of the pixel in the polarization image, where the calculation formula is: in, Represents pixel The background light intensity is in the range of , Represents pixel The original brightness value, Represents pixel The initial perspective rate; The preliminary perspective is optimized according to the local gradient value of the depth image and the pixel difference value of the difference image to obtain the final perspective, wherein the optimization formula is: in, and are the preset weight parameters for controlling the influence of local gradient and pixel difference, and Represent the maximum value of the local gradient of the depth image and the pixel difference value of the difference image, respectively. and Represents pixel points The local gradient value and pixel difference value.

[0013] Preferably, the restoration of the image content includes: For each pixel in each partition, the brightness value of the corresponding pixel in the polarized image is restored using the optimized perspective and background light intensity through the image restoration formula.

[0014] Preferably, the weighted fusion of the two restored polarization images according to the polarization degree includes: Obtaining a polarization degree difference value of each pixel point in the two polarization images, and querying a corresponding weight factor according to a predefined mapping relationship table, wherein the predefined mapping relationship table divides the polarization degree difference value into a plurality of continuous intervals, and each interval corresponds to a preset weight factor; Based on the queried weight factors, the brightness values ​​of corresponding pixels in the two restored polarization images are weightedly superimposed to generate a fused image.

[0015] The present invention also provides an underwater image enhancement system based on polarization differential imaging, which is used to perform an underwater image enhancement method based on polarization differential imaging. The system comprises: An image acquisition module is used to acquire a depth image corresponding to the underwater image to be enhanced and two polarization images whose polarization degree exceeds a set threshold, and calculate a difference image between the two polarization images; A background light intensity calculation module, used to divide the two polarization images into regions based on the difference image, and calculate the background light intensity of each divided region using the depth image; A perspective calculation module, used for calculating and optimizing the perspective of each pixel in the corresponding partition based on the background light intensity; The image enhancement module is used to restore the image content of the corresponding partition of the corresponding polarized image based on the perspective, and perform weighted fusion and global image enhancement processing on the two restored polarized images according to the polarization degree to obtain a final enhanced image.

[0016] Compared with the related art, the underwater image enhancement method and system based on polarization differential imaging provided by the present invention have the following beneficial effects: The present invention first uses a deep learning model to perform semantic segmentation on the difference image, dividing the image into a target area, a background area, and a transition area. Then, based on the information of the depth image and the difference image, the background light intensity of each partition is calculated respectively. Next, based on the background light intensity and the original brightness value of the pixel, the perspective of each pixel is calculated and optimized. Finally, the brightness value of the polarized image is restored using the optimized perspective and background light intensity, and the two restored polarized images are weighted fused and globally enhanced according to the degree of polarization to obtain the final enhanced image.

[0017] The present invention not only fully utilizes the advantages of polarization imaging technology, but also realizes more accurate enhancement of underwater images by combining information of depth images and difference images. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of an underwater image enhancement method based on polarization differential imaging provided by the present invention; Figure 2 A module structure diagram of an underwater image enhancement method based on polarization differential imaging provided by the present invention. DETAILED DESCRIPTION

[0019] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only the parts related to the present invention, rather than all structures, are shown in the accompanying drawings. In addition, the embodiments of the present invention and the features in the embodiments may be combined with each other without conflict.

[0020] It should also be noted that, for ease of description, only the parts related to the present invention, but not all of the contents, are shown in the accompanying drawings. Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the operations (or steps) as sequential processes, many of the operations therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it can also have additional steps not included in the accompanying drawings. The process can correspond to methods, functions, procedures, subroutines, subprograms, etc.

[0021] Embodiment 1 The present invention provides an underwater image enhancement method based on polarization differential imaging, referring to Figure 1 As shown, the method comprises the following steps: S1: Acquire a depth image corresponding to the underwater image to be enhanced and two polarization images whose polarization degrees exceed a set threshold, and calculate a difference image between the two polarization images.

[0022] In this embodiment, active imaging technologies including but not limited to structured light projection, binocular stereo vision or ToF (Time-of-Flight) camera are used to synchronously collect a depth image corresponding to the underwater image to be enhanced, and each pixel value in the depth image represents the distance from the corresponding point in the scene to the camera. It is necessary to ensure that the spatial resolution of the depth image is consistent with that of the polarization image, and align the coordinate system through calibration parameters.

[0023] Use an underwater camera with an adjustable polarizer, fix the shooting angle and lighting conditions, rotate the polarizer to two orthogonal directions (0° and 90°), and take two polarized images respectively. The difference in polarization degree must be greater than or equal to 30%.

[0024] The two polarization images are registered to eliminate the displacement error caused by shooting angle or movement. The SIFT feature point matching method is used to align the two images to ensure that each pixel corresponds to the same scene position in the two images.

[0025] For the two registered polarization images, the absolute value of the intensity difference is calculated pixel by pixel to generate a difference image, whose pixel values ​​reflect the local scattering difference between the two polarization images, and the high-value area corresponds to the edge of the target or the area with dense suspended particles.

[0026] S2: Based on the difference image, the two polarization images are divided into regions respectively, and the background light intensity of each divided region is calculated using the depth image.

[0027] In step S2, the area division specifically includes: First, the difference image is processed based on a deep learning semantic segmentation model to identify and mark the target area, background area and transition area.

[0028] In this embodiment, for the construction and training of the semantic segmentation model, it is necessary to collect polarization difference image datasets of underwater scenes with different turbidity, lighting conditions and target types, and manually annotate the target area (difference value ≥ 60 and matching the depth profile), background area (difference value ≤ 30 and depth variance < 5%) and transition area (difference continuously changes and is related to the depth gradient), and rotate the data, simulate light attenuation and enhance Gaussian noise to improve generalization. The model adopts an improved U-Net architecture, the encoder uses ResNet-18 to extract features, the decoder gradually restores the resolution through transposed convolution and outputs a three-channel probability map, and the activation function uses ReLU in the encoder and LeakyReLU (negative slope 0.2) in the decoder to balance the gradient stability. During the training process, the weighted cross entropy loss (target weight 1.5, background 1.0, transition 2.0) is jointly optimized with the edge-sensitive regularization term, and the training is carried out by the AdamW optimizer with cosine annealing scheduling, and the early stopping strategy prevents overfitting.

[0029] During inference, the output probability map is thresholded (target area > 0.7, background area > 0.6) and isolated areas with an area less than 50 pixels are filtered out. The edges of the transition area are then refined through dilation and erosion operations (kernel size 5×5).

[0030] Next, the marking results are mapped to the corresponding two polarization images, and the two polarization images are divided into a target area, a background area, and a transition area.

[0031] In this embodiment, the pixel-level coordinate mapping relationship between the difference image and the two polarized images is established through the pre-calibrated camera internal parameters (focal length, principal point) and external parameters (polarizer rotation angle), and then the segmentation results of the difference image (target, background, transition area mask) are mapped to the two polarized image coordinate systems through bilinear interpolation. During the mapping process, the mean difference in polarization degree of the two polarized images in the target area needs to be calculated. If the deviation from the mean difference in the target area in the difference image exceeds 15%, it is determined to be a mapping error, triggering re-registration, verifying the depth variance of the background area (needed to be <5%), otherwise adjusting the segmentation threshold. Finally, a three-region mask with the same resolution as the polarized image is output, in which the target area is used for detail recovery, the background area suppresses scattering, and the transition area achieves smooth connection.

[0032] In step S2, the background light intensity of each partition is calculated as follows: For the background area, all pixel difference values ​​belonging to the background area are extracted from the difference image, and the variance of these difference values ​​is calculated. The variance reflects the scattering uniformity of the background area: the smaller the variance, the more concentrated the difference distribution in the background area (that is, the water body scatters more uniformly), and a higher weight is given at this time.

[0033] When calculating the weight, the inverse of the variance is taken as the weight factor (if the variance is close to zero, a minimum value is added to prevent division by zero errors). The larger the weight, the higher the contribution of the corresponding depth value to the background light intensity.

[0034] The pixel values ​​of the background area in the depth image are weighted averaged according to the above weights to finally obtain the background light intensity.

[0035] For the target area, the average intensity ratio of the target area in the two polarization images is calculated to obtain the polarization degree ratio. The ratio ranges from 0 to 1 and is used to quantify the modulation ability of the target area on polarized light. The higher the ratio, the more significant the effect of reflection or scattering of the target surface on the polarization characteristics.

[0036] The polarization attenuation coefficient is derived based on the polarization degree ratio. This coefficient is used to quantify the attenuation degree of the polarization characteristics of the target surface reflection, and the value range is limited to 0.2-0.8 (to avoid physical contradictions).

[0037] The calculation of the polarization attenuation coefficient is combined with the experimentally calibrated environmental polarization interference factor, which is used to correct external interferences such as water turbidity and lighting conditions. In a controllable experimental environment (such as a clear water tank and a turbid water tank), the known reflection characteristics of a standard target (such as a grayscale plate) are measured, and the theoretical polarization ratio is compared with the actual imaging results. The environmental polarization interference factor is fitted and finally the polarization attenuation coefficient is obtained by multiplying the polarization ratio by the environmental interference factor.

[0038] The Sobel operator is used to calculate the local gradient of the depth image. The larger the gradient value, the more significant the geometric change of the target surface. The polarization attenuation coefficient is multiplied by the normalized depth gradient to generate an adaptive weight. This weight gives higher priority to geometrically complex areas (high gradient) on the target surface to ensure that the background light intensity calculation is more in line with the actual physical structure. The pixel values ​​of the target area in the depth image are weighted averaged according to the weight to obtain the background light intensity of the target area.

[0039] For the transition area, Canny edge detection is performed on the difference image (Gaussian kernel = 1.5, low / high thresholds are 50 and 150 respectively), the boundaries with significant scattering changes are marked, the pixel intensity difference of the two polarization images is calculated, and the areas where the difference is greater than 30% of the maximum difference are retained. The intersection with the edge detection result is taken to generate a candidate mask for the transition area. The mask is closed (kernel size 3×3) to fill the holes, and the intersection with the original segmentation result is taken to eliminate noise interference.

[0040] Next, for each pixel in the transition area, the Euclidean distance to the nearest background area and target area is calculated respectively. The closer the distance, the higher the background light intensity weight of the corresponding area. According to the distance weight, the background light intensity of the background and target areas is linearly interpolated to obtain the light intensity value of the transition area to ensure a smooth transition between different areas.

[0041] S3: Based on the background light intensity, calculate and optimize the perspective rate of each pixel in the corresponding partition.

[0042] In this embodiment, for each pixel in each partition, the preliminary perspective is calculated based on the background light intensity and the original brightness value of the pixel in the polarization image, where the calculation formula is: in, Represents pixel The original brightness value, Represents pixel The initial perspective is the proportion of light that is not scattered and absorbed by water when it reaches the camera from the surface of the object. Represents pixel The background light intensity is in the range of , Represents the background light intensity of the pixel, that is, the influence of ambient light caused by scattering and absorption.

[0043] In this embodiment, the numerator in the formula represents the net brightness value after removing the background light; the denominator ensures that when the background light is close to 1 (that is, the ambient light is very strong), the transmittance tends to zero, which is consistent with the actual physical reality. This formula is essentially to estimate the actual transmitted light intensity by subtracting the background light.

[0044] The preliminary perspective is optimized according to the local gradient value of the depth image and the pixel difference value of the difference image to obtain the final perspective, wherein the optimization formula is: in, and are the preset weight parameters for controlling the influence of local gradient and pixel difference, , and Represent the maximum value of the local gradient of the depth image and the pixel difference value of the difference image, respectively. and Represents pixel points The local gradient value and pixel difference value.

[0045] In this embodiment, the places with larger local gradient values ​​usually correspond to the target boundaries or areas with rich details. , which can enhance the perspective of these areas, making the restored image clearer in these places; the places where the pixel difference values ​​of the difference image are large usually correspond to the significant changes between the target and the background; by introducing The influence of noise or outliers on the perspective can be suppressed, making the restored image smoother in these places.

[0046] use and The local gradient values ​​and pixel difference values ​​are normalized to ensure that their influence is within a reasonable range and to avoid unreasonable deviations in the results caused by excessively large or small values.

[0047] The final perspective value range should be limited to If it exceeds this range, it will be truncated to ensure physical meaning.

[0048] S4: restoring the image content of the corresponding partition of the corresponding polarization image based on the perspective, and performing weighted fusion and global image enhancement processing on the two restored polarization images according to the polarization degree to obtain a final enhanced image.

[0049] In step S4, the restoration of the image content includes the following steps: For each pixel in each partition, the optimized perspective and background light intensity are used to restore the brightness value of the corresponding pixel in the polarized image through the image restoration formula, where the image restoration formula is: in, Represents pixel The restored brightness value.

[0050] In this embodiment, This step is to remove the influence of background light. Background light is the uniform background illumination caused by the scattering of light by water, which will affect the real brightness of the image.

[0051] By subtracting the background light intensity , and obtain a preliminary image without the influence of ambient light.

[0052] This section is a further correction of the preliminary results to compensate for the light attenuation due to water scattering.

[0053] Perspective Reflects the degree of light loss on the path from the object surface to the camera sensor. Higher perspective means less light loss, lower perspective means more light loss.

[0054] Divide by The operation is equivalent to "recovering" the light lost due to scattering, making the image look clearer.

[0055] Finally, add the background light intensity , to ensure that the final image still contains information about the ambient light, avoiding the image being too dark or losing its natural feel.

[0056] In step S4, the weighted fusion step specifically includes the following steps: The polarization degree difference value of each pixel point in the two polarization images is obtained, and the corresponding weight factor is queried according to a predefined mapping relationship table, wherein the predefined mapping relationship table divides the polarization degree difference value into multiple continuous intervals, and each interval corresponds to a preset weight factor.

[0057] In this embodiment, for the two restored polarization images (0° and 90° polarization direction images), the ratio of the intensity difference between the two images to the total intensity is calculated pixel by pixel to obtain a polarization degree difference value ranging from 0 to 1. This value reflects the degree of difference in polarization characteristics of the two images at the same position: the higher the difference value, the greater the influence of polarization modulation on the region (such as the edge of the target or a highly reflective surface); the lower the difference value, the more similar the polarization characteristics of the region (such as a uniformly scattered background water body).

[0058] The polarization degree difference value is divided into multiple continuous intervals (for example, 10 intervals of equal width, each interval is 0.1), where the weight allocation rule is as follows: Low difference range (0.0-0.3): Give a single image a higher weight (for example, the first image has a weight of 0.9 and the second image has a weight of 0.1) to suppress background noise.

[0059] Medium difference interval (0.3-0.6): Use progressive weights (e.g. 0.3→0.7, 0.6→0.5) to balance the complementary information of the two images.

[0060] High difference range (0.6-1.0): Evenly distribute weights (for example, 0.5) to preserve the detailed features in the dual polarization image.

[0061] Based on the queried weight factors, the brightness values ​​of corresponding pixels in the two restored polarization images are weightedly superimposed to generate a fused image.

[0062] In this embodiment, the weight value of each pixel is queried according to the predefined mapping table, and the two restored polarization images are weighted superimposed. For example, if the polarization degree difference value of a pixel is 0.7 (high difference range), the weights of the two images are both 0.5, and the brightness value of the pixel in the superimposed fused image is the average of the brightness values ​​of the two images.

[0063] If the polarization difference value of the target area is significantly higher than that of the background (for example, the difference exceeds 0.5), it is forced to be set to an equalization weight (0.5) to avoid loss of details due to segmentation errors; bilateral filtering is performed on the fused transition area (spatial kernel size 5×5, brightness difference threshold 10) to eliminate edge artifacts caused by weight jumps; dynamic range compression is performed on the fused brightness value to avoid brightness overflow due to superposition.

[0064] After weighted fusion, global image enhancement is required, including: Contrast enhancement: Contrast-constrained adaptive histogram equalization is used to divide the image into 8×8 local grids, and the contrast in each grid is limited to 2.0 to enhance details while suppressing noise amplification.

[0065] Color restoration: Extract the chromaticity channel of the original polarized image and combine it with the fused brightness channel to restore the natural color of the underwater scene.

[0066] And according to the pre-calibrated white balance parameters, adjust the gain of the chromaticity channel to correct the blue-green cast caused by water.

[0067] Embodiment 2 The present invention also provides an underwater image enhancement system based on polarization differential imaging, which is used to perform an underwater image enhancement method based on polarization differential imaging, referring to Figure 2 As shown, the system comprises: The image acquisition module 100 is used to acquire a depth image corresponding to the underwater image to be enhanced and two polarization images whose polarization degree exceeds a set threshold, and calculate a difference image between the two polarization images; A background light intensity calculation module 200, configured to divide the two polarization images into regions based on the difference image, and calculate the background light intensity of each divided region using the depth image; The perspective calculation module 300 is used to calculate and optimize the perspective of each pixel in the corresponding partition based on the background light intensity.

[0068] The image enhancement module 400 is used to restore the image content of the corresponding partition of the corresponding polarization image based on the perspective, and perform weighted fusion and global image enhancement processing on the two restored polarization images according to the polarization degree to obtain a final enhanced image.

[0069] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0070] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, the storage medium including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically-erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0071] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. An underwater image enhancement method based on polarization differential imaging, characterized in that: The method comprises the following steps: Acquire a depth image corresponding to the underwater image to be enhanced and two polarization images whose polarization degree exceeds a set threshold, and calculate a difference image between the two polarization images; Based on the difference image, the two polarization images are divided into regions respectively, and the background light intensity of each divided region is calculated using the depth image; Based on the background light intensity, calculating and optimizing the perspective rate of each pixel in the corresponding partition; The image content of the corresponding partition of the corresponding polarization image is restored based on the perspective, and the two restored polarization images are weightedly fused and globally enhanced according to the polarization degree to obtain a final enhanced image.

2. The underwater image enhancement method based on polarization differential imaging according to claim 1, characterized in that: The partitions include a target area, a background area, and a transition area.

3. The underwater image enhancement method based on polarization differential imaging according to claim 1, characterized in that: Polarization image acquisition includes: Two polarization images are acquired in sequence at different polarization degrees using the same shooting device, and the polarization degree difference exceeds a set threshold.

4. The underwater image enhancement method based on polarization differential imaging according to claim 1, characterized in that: The acquisition of the difference image includes: For two polarization images, pixel-level matching is performed at the same spatial position; For each pair of matching pixels, the difference between the pixel values ​​of the two polarization images is calculated and the absolute value is taken to obtain a difference image.

5. The underwater image enhancement method based on polarization differential imaging according to claim 2, characterized in that: The step of dividing the two polarization images into regions based on the difference image comprises: Processing the difference image based on a deep learning semantic segmentation model to identify and mark the target area, background area and transition area; The marking results are mapped to the corresponding two polarization images, and the two polarization images are divided into a target area, a background area, and a transition area.

6. The underwater image enhancement method based on polarization differential imaging according to claim 5, characterized in that: The step of calculating the background light intensity of each partition after the division using the depth image comprises: For the background area, the pixel difference values ​​of the corresponding area in the difference image are extracted, the variance of the pixel difference values ​​is calculated and used as a weight factor, and the pixel values ​​of the depth image are weighted averaged to obtain the background light intensity of the background area; For the target area, the polarization attenuation coefficient is determined according to the polarization degree ratio of the two polarization images in the target area. Combined with the local gradient information of the depth image, the adaptive weight is generated through the normalized product function to calculate the background light intensity of the target area. For the transition area, edge detection is performed on the difference image and a threshold comparison is performed with the pixel intensity difference of the two polarization images to generate a mask. Based on the mask, bidirectional linear interpolation is performed on the background light intensity of the background area and the target area to obtain the background light intensity of the transition area.

7. The underwater image enhancement method based on polarization differential imaging according to claim 6, characterized in that: The calculating and optimizing the perspective rate of each pixel in the corresponding partition based on the background light intensity includes: For each pixel in each partition, the preliminary perspective is calculated based on the background light intensity and the original brightness value of the pixel in the polarization image, where the calculation formula is: in, Represents pixel The background light intensity is in the range of , Represents pixel The original brightness value, Represents pixel The initial perspective rate; The preliminary perspective is optimized according to the local gradient value of the depth image and the pixel difference value of the difference image to obtain the final perspective, wherein the optimization formula is: in, and are the preset weight parameters for controlling the influence of local gradient and pixel difference, and Represent the maximum value of the local gradient of the depth image and the pixel difference value of the difference image, respectively. and Represents pixel points The local gradient value and pixel difference value.

8. The underwater image enhancement method based on polarization differential imaging according to claim 7, characterized in that: The restoration of the image content includes: For each pixel in each partition, the brightness value of the corresponding pixel in the polarized image is restored using the optimized perspective and background light intensity through the image restoration formula.

9. The underwater image enhancement method based on polarization differential imaging according to claim 7, characterized in that: The weighted fusion of the two restored polarization images according to the polarization degree includes: Obtaining a polarization degree difference value of each pixel point in the two polarization images, and querying a corresponding weight factor according to a predefined mapping relationship table, wherein the predefined mapping relationship table divides the polarization degree difference value into a plurality of continuous intervals, and each interval corresponds to a preset weight factor; Based on the queried weight factors, the brightness values ​​of corresponding pixels in the two restored polarization images are weightedly superimposed to generate a fused image.

10. An underwater image enhancement system based on polarization differential imaging, used to execute the underwater image enhancement method based on polarization differential imaging as claimed in any one of claims 1 to 9, characterized in that: The system comprises: An image acquisition module is used to acquire a depth image corresponding to the underwater image to be enhanced and two polarization images whose polarization degree exceeds a set threshold, and calculate a difference image between the two polarization images; A background light intensity calculation module, used to divide the two polarization images into regions based on the difference image, and calculate the background light intensity of each divided region using the depth image; A perspective calculation module, used for calculating and optimizing the perspective of each pixel in the corresponding partition based on the background light intensity; The image enhancement module is used to restore the image content of the corresponding partition of the corresponding polarized image based on the perspective, and perform weighted fusion and global image enhancement processing on the two restored polarized images according to the polarization degree to obtain a final enhanced image.

Images