An underwater polarization imaging method based on stokes vector anisotropic diffusion
By constructing a Stokes vector anisotropic diffusion model, acquiring and processing multiple underwater polarization images, the problem of distinguishing between the background and target areas in underwater imaging was solved, achieving high-quality underwater imaging results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGSU UNIV OF SCI & TECH
- Filing Date
- 2026-03-19
- Publication Date
- 2026-06-05
AI Technical Summary
Existing underwater polarization imaging technology struggles to effectively distinguish between background and target areas in complex underwater environments with high turbidity and strong noise, leading to degraded image quality, particularly in the estimation of background scattered light and poor descattering performance.
By constructing a Stokes vector anisotropic diffusion model, acquiring multiple underwater images with different polarization directions, performing brightness normalization processing, calculating gradient differences, introducing the Perona-Malik anisotropic diffusion model, constructing a Stokes vector diffusion model, estimating backscattered light, and performing depolarization compensation, a clear imaging result was finally obtained.
It effectively distinguishes between the background area and the target edge area, improving the robustness and detail preservation of underwater imaging. It is suitable for targets with complex polarization characteristics and significantly improves the imaging quality.
Smart Images
Figure CN122156003A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of underwater polarization imaging technology, specifically relating to an underwater polarization imaging method based on Stokes vector anisotropic diffusion. Background Technology
[0002] The execution of underwater vision tasks is constrained by both the absorption effect of the water itself and the scattering effect of turbid media, which can easily lead to problems such as reduced image contrast and blurred details. Among these, the "fog" phenomenon formed by backscattered light is the main cause of image quality degradation. Scattering and absorption caused by suspended particles can significantly restrict visual information perception and accurate target identification.
[0003] Current underwater polarization imaging techniques still face bottlenecks such as biased background scattering estimation and poor descattering performance. Especially in complex underwater environments with high turbidity and strong noise, existing methods struggle to effectively distinguish between background and target areas, thus affecting the quality of the final reconstructed image. Summary of the Invention
[0004] Purpose of the invention: To address issues such as inaccurate background scattering estimation, underwater scattering, and noise, this invention proposes an underwater polarization imaging method based on Stokes vector anisotropic diffusion. By constructing a Stokes vector diffusion model and performing Stokes vector diffusion processing, the background region and target region are effectively separated, and the background scattering light is accurately estimated, thereby improving the final imaging effect.
[0005] Technical solution: An underwater polarization imaging method based on Stokes vector anisotropic diffusion, comprising the following steps:
[0006] Step 1: Acquire multiple underwater polarization images with different polarization directions, perform brightness normalization processing on the acquired underwater polarization images, and construct Stokes vectors based on the brightness-normalized underwater polarization images;
[0007] Step 2: The diffusion process of the Stokes vector in different regions is approximated by calculating the gradient difference in four directions, and the required gradient in different regions is estimated accordingly; the Perona-Malik anisotropic diffusion model is introduced, and based on the gradient, the Stokes vector diffusion model is constructed by Euler discrete approximation to obtain the diffused Stokes vector.
[0008] Step 3: Calculate the polarization angle and degree of polarization based on the diffused Stokes vector; perform depolarization compensation correction on the degree of polarization to obtain the corrected degree of polarization; estimate the backscattered light based on the polarization angle and the corrected degree of polarization.
[0009] Step 4: Based on the estimated backscattered light, the final imaging result is obtained.
[0010] Furthermore, the method of approximating the diffusion process of the Stokes vector in different regions by calculating the gradient difference in four directions, and thereby estimating the required gradient in different regions, includes:
[0011] Define multiple underwater polarization images including , , , ; , , , This is the underwater polarization image after brightness normalization;
[0012] The Stokes vector , is represented as:
[0013]
[0014] Where m=0, The total light intensity of the underwater image; when m=1, The difference in light intensity is between the 0° and 90° directions of the image; when m=2, The difference in light intensity at 45° and 135°;
[0015] The gradient required for different regions is expressed as
[0016]
[0017] in, Let (x, y) be the position of the Stokes vector at that pixel. , , , These represent the differences between the current pixel and its neighboring pixels in the north, south, east, and west directions, respectively. Represents the gradient. Represents the Stokes vector The gradient magnitude at that pixel.
[0018] Furthermore, the Perona-Malik anisotropic diffusion model is introduced. Based on the gradient, a Stokes vector diffusion model is constructed using the Euler discrete approximation to obtain the diffused Stokes vector, expressed as:
[0019]
[0020] In the formula, The Stokes vector after diffusion. , through vector For time step The continuous partial differential process describes how the Stokes vector changes continuously with iteration time. For divergence operators, Indicates the diffusion coefficient. Represents image gradient; diffusion coefficient With gradient magnitude and gradient threshold parameters Related; after discretizing the partial differential equation After updating the Euler discrete form, iterate. Stokes vector of the order of 2, The time step set for the discrete form.
[0021] Furthermore, based on the diffused Stokes vector, the polarization angle and degree of polarization are calculated and expressed as:
[0022]
[0023] in, For degree of polarization, It is the polarization angle. and For process parameters.
[0024] Furthermore, the polarization degree is depolarized and compensated to obtain the corrected polarization degree, which is expressed as:
[0025]
[0026] in, This is the corrected degree of polarization. This refers to the optical thickness or the equivalent number of scattering events. It is a constant related to the properties of the medium.
[0027] Furthermore, the estimation of backscattered light based on the polarization angle and the corrected degree of polarization is expressed as follows:
[0028]
[0029] In the formula, , Indicates the backscattered light component. Represents the darkest image. This represents backscattered light.
[0030] Furthermore, the process of obtaining the final imaging result based on the estimated backscattered light includes:
[0031] The estimated backscattered light and transmittance are combined and substituted into a traditional underwater imaging physical model to obtain the final imaging result, expressed as follows:
[0032]
[0033] in, To set the total light intensity, the total light intensity of the underwater image is... Gaussian filtering was performed to obtain... This represents the standard deviation of the Gaussian filter. The final imaging result is given, where t represents the transmittance. The intensity of backscattered light at infinity.
[0034] Beneficial Effects: This invention acquires images at different polarization angles and performs brightness normalization processing to obtain an initial Stokes vector. Then, a Perona-Malik anisotropic diffusion model is introduced to construct a Stokes vector diffusion model to distinguish between background and target edge regions. Based on the diffused Stokes vector, the polarization angle and degree of polarization are calculated, and multiple scattering is considered to perform depolarization compensation processing on the degree of polarization. Based on this, backscattered light is estimated. Finally, the backscattered light and transmittance are substituted into a traditional underwater imaging physical model to obtain the final restored image. Compared with existing technologies, this invention has the following advantages:
[0035] (1) The present invention constructs Stokes vectors by multi-angle polarization images and uses anisotropic diffusion to construct a Stokes diffusion model to effectively distinguish the background region of the Stokes vector from the target edge region;
[0036] (2) This invention introduces multiple scattering to depolarize the degree of polarization compensation, and at the same time accurately estimates the backscattered light, thus realizing the effective restoration of images in underwater environments with different turbidity.
[0037] (3) The present invention can significantly suppress background interference and has good adaptability to targets with complex polarization characteristics, thereby effectively improving the robustness, detail preservation ability and applicability of underwater polarization imaging, and has important engineering application value. Attached Figure Description
[0038] Figure 1 This is a flowchart of the present invention;
[0039] Figure 2 This is a model diagram of the underwater active polarization imaging system built according to the present invention;
[0040] Figure 3 The diagram illustrates the actual effect of applying the method of this invention. Figure 3 Images (a) and (c) in the image are the total light intensity images underwater. Figure 3 (b) and (d) are illustrations showing the effects of the actual application of the present invention. Detailed Implementation
[0041] To make the objectives, technical solutions, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0042] like Figure 1 As shown, this invention proposes an underwater polarization imaging method based on Stokes vector anisotropic diffusion, which mainly includes the following steps:
[0043] Step 1: Acquire four underwater images with different polarization directions. , , , The underwater image is normalized for brightness to obtain the normalized Stokes vector of the underwater polarization image. Specific operations include:
[0044] use Figure 2 The underwater active polarization imaging simulation system shown has an LED red light source 1 emitting a starting beam that passes sequentially through a transparent water tank 3 and illuminates suspended particles 4 and a target object 5. 20 mL of skim milk is added to the clear water to simulate a turbid medium. The target object 5 can be a metal coin with high polarization characteristics or a plastic toy with low polarization characteristics, used to characterize the imaging differences in surface materials and polarization properties. When the starting beam illuminates the target object 5, the target object 5 reflects the light to form the target beam, and the suspended particles 4 generate backscattered light. Both beams enter the polarization camera 7, which acquires underwater polarization images in different polarization directions for subsequent Stokes vector construction and image restoration processing.
[0045] The total light intensity received by the detector has the following relationship with the target light and the backscattered light:
[0046]
[0047] in, This represents the total light intensity of the underwater image; For target light; t represents backscattered light, and t represents transmittance.
[0048] Four polarization images are generated by receiving data through a polarization camera as the initial input source.
[0049] The four polarization images are normalized and represented as follows:
[0050]
[0051] Here, `scaling factor` is a scaling factor that can be set using the median brightness of the image to ensure consistent brightness across all images; `median()` calculates the median of pixel values in the image to remove outliers. Because underwater images may sometimes contain unusual highlights or shadows, the median method avoids the impact of these extreme values on normalization; `max()` selects the maximum value or maximum pixel value (or brightness value) in the image for scaling during normalization.
[0052] Based on the normalized polarization image, a Stokes vector is constructed, represented as:
[0053]
[0054] Among them, the normalized Stokes vector middle, This represents the total light intensity of the underwater image; The difference in light intensity at 0° and 90° in the image; The difference in light intensity is between the 45° and 135° directions.
[0055] Step 2: Introduce the Perona-Malik anisotropic diffusion model and construct a Stokes vector diffusion model to distinguish between the background region and the target edge region. Specific operations include:
[0056] Because the gradient changes are smaller in the background region and larger in the target edge region, the Stokes vector requires different processing during diffusion. An anisotropic diffusion model is used to differentiate regions based on their varying gradient responses: in the background region, a larger diffusion intensity is determined by the diffusion coefficient, effectively smoothing the background changes; in the target edge region, a smaller diffusion intensity is determined, avoiding the loss of target edge information. Based on this characteristic, the Stokes vector can be considered to have different diffusion characteristics in different regions. Its diffusion process is approximated by calculating the gradient differences in four directions, and the required gradient magnitude is estimated accordingly. This processing effectively distinguishes the background region from the target edge region, and the Stokes vector can be processed using the diffusion coefficient and gradient pair. Adjustments:
[0057]
[0058] in, Let (x, y) be the position of the Stokes vector at that pixel. , , , These represent the differences between the current pixel and its neighboring pixels in the north, south, east, and west directions, respectively. Gradient Describe the direction and rate of change of the Stokes vector. Stokes vector The gradient magnitude at that point reflects the intensity of change in the image at that point.
[0059] Introducing the Perona-Malik anisotropic diffusion model, a Stokes vector diffusion model is constructed using the Euler discrete approximation:
[0060]
[0061] In the formula, The Stokes vector after diffusion. . This is a divergence operator used to calculate the change in diffusion at each pixel and incorporate these changes into the new pixel value. It is the diffusion coefficient. In flat regions, the gradient is small, the diffusion coefficient is large, and the diffusion is strong; in edge regions, the gradient is large, the diffusion coefficient is small, and the diffusion is weak. Whether edge information can be preserved depends on the gradient magnitude. Calculate whether diffusion should be suppressed, if If the value is too large, the diffusion coefficient tends to 0, indicating that the area is the edge of the target object and it is desirable that the diffusion not occur; conversely, if the value is too small, it is considered the background and it is desirable that the diffusion occur. This parameter controls the gradient threshold and determines the sensitivity of the diffusion. Since images are discrete, continuous partial differential equations cannot be directly applied; therefore, numerical discretization is necessary. After updating the Euler discrete form, iterate. Stokes vector of the order of 1. For a set time step, this method adjusts pixel values based on the diffusion coefficient to gradually achieve smooth processing of the Stokes vector.
[0062] Step 3: Based on the diffused Stokes vector, calculate the polarization angle and degree of polarization, and considering multiple scattering, perform depolarization compensation on the degree of polarization. Then, estimate the backscattered light. Specific calculations include:
[0063] Methods for calculating polarization angle and corrected degree of polarization:
[0064]
[0065] In highly turbid media, light often undergoes numerous scattering events during propagation. Some Monte Carlo simulations show that the surviving polarization degree decreases approximately exponentially with the number of scattering events or optical thickness; therefore, depolarization compensation correction is necessary. To compensate for the estimated corrected degree of polarization, This can be understood as optical thickness or equivalent scattering number. It is a constant related to the properties of the medium. The degree of polarization obtained before correction. It represents the polarization angle. and For process parameters.
[0066] The method for estimating backscattered light is as follows:
[0067]
[0068] In low-scattering environments, the intensity distribution trends of target light and backscattered light in the darkest polarization image are consistent. Furthermore, as the scattering effect increases, the information of the target light is almost entirely obscured by the backscattered light from the scattering medium. Therefore, the darkest image can be... Considered as backscattered light component , It is calculated from the diffused Stokes vector and polarization angle. Based on the polarization degree correlation characteristics, it can be based on... get And finally, backscattered light is obtained. .
[0069] Step 4: Combine backscattered light and transmittance, and substitute them into the traditional underwater imaging physical model to obtain the final imaging result, expressed as follows:
[0070]
[0071] in, To set the total light intensity, for Perform a simple Gaussian filter to reduce noise interference with the restoration result. The standard deviation of the Gaussian filter determines the strength of the filter. This is the final imaging result. t represents the transmittance. The intensity of backscattered light at infinity can be set as the total light intensity. The average value of the top 0.1% of pixels with the highest grayscale value in the image.
[0072] To verify the effectiveness of the present invention, experiments were conducted on target materials made of different materials, such as... Figure 3 In the images (a) and (c), a metal coin and a plastic toy, respectively, the image noise and detail are severely lost due to the scattering effect of the water. However, after processing by the method of this invention, the image clarity and visibility are significantly improved. Figure 3 (b) and (d) in the text are also applicable to underwater environments with different turbidity levels.
[0073] Reference underwater image enhancement metrics ( Image gradient () Information entropy Objective quality evaluation was performed on the restored images. A higher value indicates higher image quality. The results are shown in the table below:
[0074] Table 1. Quality evaluation of reconstructed images of metal coins and complex surface targets.
[0075]
[0076] Table 2. Quality evaluation of the restored images of targets with relatively smooth surfaces on plastic toys.
[0077]
[0078] As can be seen from the table, the restored image of this invention shows significant improvements in all objective evaluation indicators and noise suppression compared to the original total intensity image, verifying the effectiveness of this invention in restoring different target surfaces. The method of this invention can effectively distinguish between "scattered background" and "true edges," and is applicable to targets with complex polarization characteristics, significantly improving descattering and noise reduction effects.
Claims
1. An underwater polarization imaging method based on Stokes vector anisotropic diffusion, characterized in that: Includes the following steps: Step 1: Acquire multiple underwater polarization images with different polarization directions, perform brightness normalization processing on the acquired underwater polarization images, and construct Stokes vectors based on the brightness-normalized underwater polarization images; Step 2: The diffusion process of the Stokes vector in different regions is approximated by calculating the gradient difference in four directions, and the required gradient in different regions is estimated accordingly; the Perona-Malik anisotropic diffusion model is introduced, and based on the gradient, the Stokes vector diffusion model is constructed by Euler discrete approximation to obtain the diffused Stokes vector. Step 3: Calculate the polarization angle and degree of polarization based on the diffused Stokes vector; perform depolarization compensation correction on the degree of polarization to obtain the corrected degree of polarization; estimate the backscattered light based on the polarization angle and the corrected degree of polarization. Step 4: Based on the estimated backscattered light, the final imaging result is obtained.
2. The underwater polarization imaging method based on Stokes vector anisotropic diffusion according to claim 1, characterized in that: The method of approximating the diffusion process of the Stokes vector in different regions by calculating the gradient difference in four directions, and estimating the required gradient in different regions accordingly, includes: Define multiple underwater polarization images including , , , ; , , , This is the underwater polarization image after brightness normalization; The Stokes vector , is represented as: ； Where m=0, The total light intensity of the underwater image; when m=1, The difference in light intensity is between the 0° and 90° directions of the image; when m=2, The difference in light intensity at 45° and 135°; The gradient required for different regions is expressed as ； in, Let (x, y) be the position of the Stokes vector at that pixel. , , , These represent the differences between the current pixel and its neighboring pixels in the north, south, east, and west directions, respectively. Represents the gradient. Represents the Stokes vector The gradient magnitude at that pixel.
3. The underwater polarization imaging method based on Stokes vector anisotropic diffusion according to claim 2, characterized in that: The introduced Perona-Malik anisotropic diffusion model, based on the gradient, constructs a Stokes vector diffusion model using the Euler discrete approximation, yielding the diffused Stokes vector, expressed as: ； In the formula, The Stokes vector after diffusion. , through vector For time step The continuous partial differential process describes how the Stokes vector changes continuously with iteration time. For divergence operators, Indicates the diffusion coefficient. Represents image gradient; diffusion coefficient With gradient magnitude and gradient threshold parameters Related; after discretizing the partial differential equation After updating the Euler discrete form, iterate. Stokes vector of the order of 2, The time step set for the discrete form.
4. The underwater polarization imaging method based on Stokes vector anisotropic diffusion according to claim 3, characterized in that: The polarization angle and degree of polarization are calculated based on the diffused Stokes vector and expressed as follows: ； ； ； in, For degree of polarization, It is the polarization angle. and For process parameters.
5. The underwater polarization imaging method based on Stokes vector anisotropic diffusion according to claim 4, characterized in that: The polarization degree is depolarized and compensated to obtain the corrected polarization degree, which is expressed as follows: ； in, This is the corrected degree of polarization. This refers to the optical thickness or the equivalent number of scattering events. It is a constant related to the properties of the medium.
6. The underwater polarization imaging method based on Stokes vector anisotropic diffusion according to claim 5, characterized in that: The estimation of backscattered light based on the polarization angle and the corrected degree of polarization is expressed as follows: ； In the formula, , Indicates the backscattered light component. Represents the darkest image. This represents backscattered light.
7. The underwater polarization imaging method based on Stokes vector anisotropic diffusion according to claim 6, characterized in that: The final imaging result is obtained based on the estimated backscattered light, including: The estimated backscattered light and transmittance are combined and substituted into a traditional underwater imaging physical model to obtain the final imaging result, expressed as follows: ； in, To set the total light intensity, the total light intensity of the underwater image is... Gaussian filtering was performed to obtain... This represents the standard deviation of the Gaussian filter. The final imaging result is given, where t represents the transmittance. The intensity of backscattered light at infinity.