Underwater image restoration method based on polarization imaging and low-frequency filtering

Abstract

The invention discloses an underwater image restoration method based on polarization imaging and low-frequency filtering, and the method comprises the steps: S1, collecting polarization images of different turbid water bodies at three angles through an underwater polarization imaging experiment platform, and calculating a Stokes vector of an underwater target scene; s2, calculating a polarization degree and a polarization angle image of an underwater target scene according to the Stokes vector, and establishing a target light model according to a relation of polarization information; s3, based on the low-frequency characteristics of the background light, extracting low-frequency background light components of the turbid polarization image at each angle, and calculating a corresponding background light Stokes vector according to the low-frequency background light components; and S4, the processed parameters are substituted into the imaging model, and the optimal cut-off frequency of the filter is adaptively determined in combination with enhanced measurement evaluation indexes, so that high-quality underwater image restoration is realized. According to the method, texture features and edge details of more target areas can be effectively restored.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an underwater image restoration method based on polarization imaging and low-frequency filtering. Background Art

[0002] The development of underwater detection technology has made visual perception a core support for deep-sea development, ecological monitoring, and military security. However, water absorption, scattering, and light attenuation result in common problems in imaging such as color distortion, reduced contrast, and blurred details, severely limiting target recognition and operation efficiency. Solving the problem of imaging quality attenuation is the core challenge and key direction for enhancing marine exploration capabilities and promoting the intelligent development of underwater equipment.

[0003] Underwater polarization imaging technology can effectively suppress the interference of water scattering light by extracting the polarization feature differences between the target and the background, significantly enhancing the contrast of underwater images and the target recognition ability, and has become an effective means to solve the problem of underwater vision degradation. Current research mainly focuses on the extraction and modeling of polarization degree (DoP) information, and realizes scattering suppression by establishing a Stokes vector calculation model or developing a polarization difference imaging algorithm. However, these methods generally have the defect of insufficient utilization of polarization angle (AoP) information, resulting in the under-exploitation of the AoP distribution characteristics of background light and limited imaging stability under turbid water or complex lighting conditions. Therefore, how to effectively explore the mechanism of the role of AoP information in water scattering suppression and systematically integrate it into the image restoration model has become the core problem that needs to be broken through to improve the quality of underwater polarization imaging. Summary of the Invention

[0004] The purpose of the present invention is to provide an underwater image restoration method based on polarization imaging and low-frequency filtering to solve the problems mentioned in the background art.

[0005] To achieve the above purpose, the present invention provides an underwater image restoration method based on polarization imaging and low-frequency filtering, including the following steps: S1. Use an underwater polarization imaging experimental platform to collect polarization images at three angles of different turbid waters, and calculate the Stokes vector of the underwater target scene; S2. Calculate the polarization degree and polarization angle images of the underwater target scene according to the Stokes vector, and establish a target light model based on the right triangle relationship of polarization information; S3. Based on the low-frequency characteristics of the background light, use a frequency-domain Gaussian low-pass filter to extract the low-frequency background light components of each angle of turbid polarization images respectively, and calculate the corresponding background light Stokes vector accordingly; S4. Substitute the processed parameters into the imaging model, and adaptively determine the optimal cut-off frequency of the filter in combination with the enhanced measurement evaluation index to achieve high-quality underwater image restoration.

[0006] Preferably, in S1, the Stokes vector I , Q and U are calculated as follows: ; where I 0 , I 60 and I 120 respectively represent the polarization images of three angles of different turbid water bodies.

[0007] Preferably, the polarization information of the underwater target scene is calculated through the Stokes vectors I , Q and U : ; where P represents the degree of polarized light in the image, θ represents the polarization direction of the polarized light in the image, represents the degree of polarization, represents the polarization angle.

[0008] Preferably, S2 includes: S21. By analyzing the polarization information, the corresponding right triangle relationship of the polarization information is as follows: The hypotenuse corresponds to the intensity PI of the polarized light part in the image, and the two right sides respectively correspond to Q and U components. Among them, the angle formed by the intensity PI of the polarized light part in the image and the Q component is 2 θ ; S22. According to the properties of a right triangle, the height PI perpendicular to the side is: ; S23. Based on the derivation of the polarization information and the height perpendicular to the PI side, the mathematical expression of the background light B is: ; where is the polarization height of B , P B is the degree of polarization of B , α is the polarization angle of B ; S24, Underwater image I is regarded as a clear target image S and the background light causing image blurring B are added together to obtain: ; S25, Combine I the calculation formula derivation to construct an underwater target light model, and obtain the solution of the clear target image S : .

[0009] Preferably, the S3 includes: S31, Based on the low-frequency characteristics of the background light, through a frequency-domain Gaussian low-pass filter, extract the corresponding low-frequency background light components from the input images I 0 , I 60 and I 120 ; ; S32, Based on the extracted background light images B 0 , B 60 and B 120 derive the Stokes vector components of the background light B , Q B and U B ; S33, Through the simultaneous operation of the polarization information and the formula of the height perpendicular to the PI edge, determine the numerical solutions of all unknown background light parameters , P B and α ; S34, Substitute the calculated background light Stokes vector into the target light model, perform parameter estimation and model inversion to obtain a preliminary restored image; S35, Use the enhanced measurement evaluation value as the image quality evaluation standard, quantitatively evaluate the quality of the restored image by calculating the enhanced measurement evaluation values at different cut-off frequencies, and screen the parameter combination that makes the enhanced measurement evaluation value reach the maximum: ; wherein, represents the cut-off frequency when the enhanced measurement evaluation value is the largest; S36, Update and iterate the cut-off frequency σ .

[0010] Preferably, the low-frequency background light component is: ; wherein, represents three different angles, taking 0, 60, and 120, F represents Fourier transform, F -1 is the inverse Fourier transform, represents a convolution operation, represents a Gaussian high-pass filter in the frequency domain; ; wherein, represents the frequency domain coordinates, σ is the cut-off frequency of the filter.

[0011] Preferably, the S36 is specifically: repeating steps S31 - S35, and recording the enhancement measurement evaluation value of each group of restoration results and the corresponding cut-off frequency.

[0012] Preferably, the S4 includes: S41. After completing the update iteration of the cut-off frequency σ , save the EME values of all restored images and the corresponding cut-off frequencies; S42. Extract the cut-off frequency σ when the EME value is the largest, and use this cut-off frequency to execute steps S31 - S34 to obtain the optimal restoration result.

[0013] Therefore, the present invention adopts the above-mentioned underwater image restoration method based on polarization imaging and low-frequency filtering, and has the following beneficial effects: (1) The present invention constructs a new target light model based on the basic theory of underwater polarization imaging technology; by utilizing the low-frequency distribution characteristics of background light, adaptive frequency domain Gaussian low-pass filtering is used to achieve background light separation and model parameter estimation, and finally underwater polarization image restoration is completed through parameter inversion.

[0014] (2) On the premise of maintaining the overall stability of the image, the present invention can effectively restore the texture features and edge details of more target areas; this detail restoration ability has practical value for improving the task performance of machine vision application scenarios such as underwater target recognition and robot path planning.

[0015] (3) To improve the utilization rate of polarization information, based on the right triangle relationship between the Stokes vector and DoP and AoP, the HoP parameter is introduced to obtain the background light expression of multi-polarization information; combined with the underwater imaging model, the solution of the target light is obtained.

[0016] (4) To accurately estimate the parameters in the target light model, an adaptive frequency-domain Gaussian low-pass filtering method is implemented using the EME value to extract the background light components of polarized images at different angles for calculating the unknown parameters in the model.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Brief Description of the Drawings

[0018] Figure 1 is a flowchart of an underwater image restoration method based on polarization imaging and low-frequency filtering according to an embodiment of the present invention; Figure 2 is a multi-angle underwater polarization map according to an embodiment of the present invention; Figure 3 is an image of DoP and AoP according to an embodiment of the present invention; Figure 4 is a schematic diagram of the right triangle relationship of polarization information according to an embodiment of the present invention; Figure 5 is a flowchart of step S4 according to an embodiment of the present invention; Figure 6 is a comparison chart of the restoration results of each method; (a) represents the underwater image I , (b) represents the CLAHE method, (c) represents the UDCP method, (d) represents the UDPLC method, and (e) represents the method of the present invention. Detailed Embodiments

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. The components of the embodiments of the present invention usually described and illustrated herein can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present invention provided herein is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0020] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0021] Embodiment As Figure 1 shown, the present invention provides an underwater image restoration method based on polarization imaging and low-frequency filtering, including the following steps: S1. Acquisition of underwater turbid polarization images, specifically: Using an underwater polarization imaging experimental platform to collect polarization images at three angles of different turbid water bodies, which are respectively represented as I 0 , I 60 and I 120 , as shown in Figure 2 , and calculate the Stokes vector of the underwater target scene I , Q and U as follows: ; Calculate the polarization information of the underwater target scene through the Stokes vector: ; Among them, P represents the degree of polarized light in the image, that is, the amplitude information, θ represents the polarization direction of the polarized light in the image, that is, the phase information.

[0022] S2. Underwater imaging modeling of AoP information, specifically: Referring to Figure 3 , calculate the degree of polarization DoP and the angle of polarization AoP images of the underwater target scene according to the Stokes vector, and establish a target light model based on the right triangle relationship of the polarization information.

[0023] In this embodiment, step S2 includes: S21. Combining the formula analysis of polarization information, the right triangle relationship of polarization information can be corresponding to the right triangle structure shown in Figure 4 : The hypotenuse corresponds to the intensity of the polarized light part in the image PI , and the two right sides respectively correspond to Q and U components, where PI and Q The included angle formed by the components is 2 θ .

[0024] S22. According to the properties of the right triangle, the height PI perpendicular to the side can be obtained as: ; S23. By introducing an intermediate variable , which is called the height of polarization (HoP) of the image here, the parameters P , θ and the light intensity I can be directly related, so as to supplement richer polarization feature information for the imaging model. Based on the polarization information and perpendicular toPI Derivation of the formula for the height of the side yields the background light of B The mathematical expression is: ; Among them, is B the degree of polarization HoP of P B is B the DoP of α is B the AoP of

[0025] S24, underwater image I can be regarded as composed of a clear target image S and the background light that causes image blurring B added together, resulting in: ; S25. Combining the calculation formula derivation of I can construct an underwater target light model, and finally obtain the solution of the clear target image S : ; Only need to estimate , P B and α and other unknown parameters, and the restoration task of the underwater polarization image can be achieved through the above formula.

[0026] S3. Adaptive frequency-domain Gaussian low-pass filtering, specifically: Based on the low-frequency characteristics of the background light, use a frequency-domain Gaussian low-pass filter to extract the low-frequency background light components of each-angle turbid polarization image, which are respectively represented as B 0 , B 60 and B 120 , and calculate the corresponding background light Stokes vector accordingly.

[0027] In this embodiment, step S3 includes: S31. Based on the low-frequency characteristics of the background light, use frequency-domain Gaussian low-pass filtering technology to extract the corresponding low-frequency background light components from the input images I 0 , I 60 and I 120 : : ; Among them, represents three different angles, Take 0, 60, and 120, F denotes the Fourier transform, F -1 is the inverse Fourier transform, represents the convolution operation, is expressed as a Gaussian high-pass filter in the frequency domain.

[0028] ; where, denotes the frequency-domain coordinates, σ is the cut-off frequency of the filter.

[0029] S32. Based on the extracted background light image B 0 、 B 60 and B 120 the Stokes vector components of the background light can be further derived B 、 Q B and U B . The input images I 0 、 I 60 and I 120 after passing through the frequency-domain Gaussian low-pass filter, we get B 0 、 B 60 and B 120 . Substituting B 0 、 B 60 and B 120 into the Stokes vector calculation formula, we obtain B 、 Q B and U B .

[0030] S33. By jointly operating on the polarization information and the formula of the height perpendicular to the PI side, the numerical solutions of all unknown background light parameters 、 P B and α are determined.

[0031] S34. Substitute the calculated background light Stokes vector into the target light model, that is, substitute the clear target image SIn the formula, parameter estimation and model inversion are performed to obtain a preliminary restored image, that is, a clear target image S 。

[0032] S35. In practical applications σ The selection of directly affects the accuracy of background light separation. To achieve σ automatic optimization of the value, the enhanced measurement evaluation (EME) value is used as the image quality evaluation criterion. The quality of the restored image is quantitatively evaluated by calculating the EME values at different cut-off frequencies, and the parameter combination that maximizes this index is selected: ; Among them represents the cut-off frequency when the enhanced measurement evaluation value is the largest.

[0033] S36. Update and iterate the cut-off frequency σ specifically: repeat steps S31 - S35, and record the EME value of each group of restored results and the corresponding cut-off frequency.

[0034] S4. Underwater polarization image restoration, specifically: by substituting the processed parameters into the imaging model, adaptively determine the optimal cut-off frequency of the filter in combination with the enhanced measurement evaluation index, and finally achieve high-quality underwater image restoration, referring to Figure 5 。

[0035] In this embodiment, step S4 includes: S41. After completing the update and iteration of the cut-off frequency σ save the EME values of all restored images and the corresponding cut-off frequencies.

[0036] S42. Extract the cut-off frequency σ when the EME value is the largest, and use this cut-off frequency to execute steps S31 - S34 to obtain the optimal restored result.

[0037] To comprehensively evaluate the performance of this method, two types of target polarization images in different turbid water body environments are selected. Based on evaluation indexes such as SSIM, PSNR, PCQI, LIPIPS, UIQM, and UCIQE, a comparative experiment is carried out between this method (OUR) and three processing methods, CLAHE, UDCP, and UDPLC. Figure 6 shows the visual comparison of the processing effects of different methods, and Table 1 provides the quantitative analysis results of the experimental data. The specific numerical values reflect the objective performance differences of each algorithm.

[0038] Table 1 Quantitative evaluation results ;

[0039] Therefore, the present invention adopts the above-mentioned underwater image restoration method based on polarization imaging and low-frequency filtering. This method optimizes the imaging model by fusing AoP information and combines the parameter adaptability of frequency-domain filtering, significantly improving the contrast and clarity of the restored image, and further enhancing the comprehensive performance of other underwater vision tasks.

[0040] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An underwater image restoration method based on polarization imaging and low-frequency filtering, characterized in that, It includes the following steps: S1. Use an underwater polarization imaging experimental platform to collect polarization images at three angles of different turbid waters, and calculate the Stokes vector of the underwater target scene; S2. Calculate the polarization degree and polarization angle images of the underwater target scene according to the Stokes vector, and establish a target light model based on the right triangle relationship of polarization information; S3. Based on the low-frequency characteristics of the background light, use a frequency-domain Gaussian low-pass filter to extract the low-frequency background light components of the turbid polarization images at each angle, and calculate the corresponding background light Stokes vector accordingly; S4. Substitute the processed parameters into the imaging model, and adaptively determine the optimal cut-off frequency of the filter in combination with the enhanced measurement evaluation index to achieve high-quality underwater image restoration.

2. The underwater image restoration method based on polarization imaging and low-frequency filtering according to claim 1, characterized in that: The Stokes vector in S1 I , Q and U are calculated as follows: ； Among them, I 0 , I 60 and I 120 respectively represent the polarization images at three angles of different turbid water bodies.

3. The underwater image restoration method based on polarization imaging and low-frequency filtering according to claim 2, wherein, By means of the Stokes vector I , Q and U calculate the polarization information of the underwater target scene: ； Among them, P represents the degree of polarized light in the image, θ represents the polarization direction of polarized light in the image, represents the degree of polarization, represents the polarization angle.

4. A method for underwater image restoration based on polarization imaging and low-frequency filtering according to claim 3, characterized in that, The S2 includes: S21. By analyzing in combination with polarization information, the right triangle relationship of the polarization information is obtained as follows: the hypotenuse corresponds to the intensity of the polarized light part in the image PI , and the two right sides respectively correspond to Q and U components, where the intensity of the polarized light part in the image PI and Q components form an angle of 2 θ ; S22. According to the properties of a right triangle, the height perpendicular to PI the side is: ； S23. Derivation of the ambient light based on the polarization information and the height of the side perpendicular to the polarized light part in the image PI yields the mathematical expression of the ambient light as follows: B ​ ； Among them, is B with high polarization, P B is B the degree of polarization of α is B the polarization angle of; S24, Underwater Image I Regarded as a clear target image S and the background light causing image blurring B are added together to obtain: ； S25. Combine I to derive the underwater target optical model based on the calculation formula, and obtain a clear target image S Solution: 。 5. The underwater image restoration method based on polarization imaging and low-frequency filtering according to claim 4, characterized in that, The S3 includes: S31, based on the low-frequency characteristics of background light, through the frequency domain Gaussian low-pass filter, from the input image I 0 , I 60 and I 120 Extract the corresponding low-frequency background light component ; S32, based on the extracted background light image B 0 , B 60 and B 120 Derivation of the Stokes vector components of the background light B , Q B and U B ; S33. By performing a combined operation of the polarization information and the formula for the height perpendicular to the PI side, determine the numerical solutions of all unknown background light parameters , P B and α ; S34. Substitute the calculated background light Stokes vector into the target light model for parameter estimation and model inversion to obtain a preliminary restored image; S35. Use the enhanced measurement evaluation value as the image quality evaluation standard, and quantitatively evaluate the quality of the restored image by calculating the enhanced measurement evaluation values at different cut-off frequencies, and screen the parameter combination that makes the enhanced measurement evaluation value reach the maximum value: ； Among them, represents the cut-off frequency when the enhanced measurement evaluation value is the largest; S36. Update and iterate the cut-off frequency σ ​ 6. The underwater image restoration method based on polarization imaging and low-frequency filtering according to claim 5, characterized in that The low-frequency background light component is as follows: ； Among them, represents three different angles, taking 0, 60, and 120, F represents the Fourier transform, F -1 is the inverse Fourier transform, represents the convolution operation, is expressed as a Gaussian high-pass filter in the frequency domain; ； Among them, represents the frequency-domain coordinate, σ is the cut-off frequency of the filter.

7. A method for underwater image restoration based on polarization imaging and low-frequency filtering according to claim 6, characterized in that The S36 is specifically: Repeat steps S31 - S35, and record the enhanced measurement evaluation values of each group of restoration results and the corresponding cut-off frequencies.

8. A method for underwater image restoration based on polarization imaging and low-frequency filtering according to claim 7, characterized in that, The S4 includes: S41. Complete the update iteration of the cut-off frequency σ After that, save the EME values of all restored images and the corresponding cut-off frequencies; S42. Extract the cut-off frequency when the EME value is the largest σ , and use this cut-off frequency to execute steps S31 - S34 to obtain the optimal restoration result.

Images