An underwater polarization imaging method based on frequency domain processing and global estimation

Abstract

This invention belongs to the field of image enhancement technology, and specifically to an underwater polarization imaging method based on frequency domain processing combined with global estimation. The invention first preprocesses orthogonal polarization images by histogram stretching and high-pass filtering to reduce the influence of low-frequency scattered light, and obtains orthogonal polarization image pairs more accurately through polarization degree correlation. Subsequently, the intensity image is processed by a frequency domain low-pass filter, and combined with an optimal threshold factor, the polarization degree of backscattered light is accurately estimated. Simultaneously, an enhancement metric (EME) is used to evaluate the descattering effect to optimize the estimation of the target light's polarization degree. Finally, image restoration is performed using a physical model. Experimental results show that the restored static targets are clearly distinguishable, and the corresponding image evaluation metrics are greatly improved. In particular, moving targets under non-uniform lighting conditions can also be clearly restored.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image enhancement, in particular to an underwater polarization imaging method based on frequency domain processing combined with global estimation. BACKGROUND

[0002] Affected by the water body itself and the turbid medium such as suspended particles and soluble organic matter in it, the light containing target information is absorbed and strongly scattered by the turbid medium, resulting in problems such as increased noise and greatly reduced contrast of the image obtained underwater. Under the scattering effect, the interference of backscattered light is the main reason for the decline in image quality. Therefore, it is worth studying to overcome the influence of backscattered light on underwater optical imaging. Since the backscattered light is partially polarized light, the underwater polarization imaging method is proposed as an effective underwater optical imaging technology to suppress backscattered light.

[0003] The previous polarization imaging despeckling method usually needs prior information of the background area, and rarely considers the influence of non-uniformity of the light field on image recovery, which not only reduces the processing speed of imaging, but also introduces errors in image recovery, especially for moving targets in complex scattering environments. Therefore, we propose an underwater polarization imaging method based on frequency domain processing combined with global estimation to solve the above problems. SUMMARY

[0004] (I) Technical problems solved

[0005] In view of the deficiencies of the prior art, the present application provides an underwater polarization imaging method based on frequency domain processing combined with global estimation, which solves the problems proposed in the background art.

[0006] (II) Technical solutions

[0007] In order to achieve the above purpose, the present application specifically adopts the following technical solutions:

[0008] An underwater polarization imaging method based on frequency domain processing combined with global estimation, comprising the following steps:

[0009] Step 1, obtain multiple underwater images with different polarization directions, and establish an underwater polarization imaging model therefrom, and further obtain the polarization degree of target reflected light and the polarization degree of backscattered light;

[0010] Step 2, first, perform histogram stretching and high-pass filtering preprocessing on the orthogonal polarization images to reduce the influence of low-frequency scattered light, and more accurately obtain the orthogonal polarization image pair through polarization degree correlation;

[0011] Step three, the intensity image is processed by a frequency domain low-pass filter, the polarization degree of the backscattering light is accurately estimated by combining the optimal threshold factor, the effect of depolarization is evaluated by using the enhancement metric to evaluate the polarization degree of the target light, and finally, the image is restored by combining the physical model.

[0012] Further, the value range of the polarization degree of the target reflected light and the backscattering light in step one is between 0 and 1.

[0013] Further, in step two, the image acquisition method of the histogram stretching and high-pass filter pretreatment of the orthogonal polarization image is: the histogram of the minimum light intensity image I ⊥ (x,y) is stretched, and the I' ⊥ (x,y) after histogram stretching is subjected to a frequency domain pretreatment operation, that is:

[0014]

[0015] Wherein I || (x,y) and I ⊥ (x,y) represent an orthogonal polarization image pair.

[0016] The stretched I' ⊥ (x,y) is subjected to Fourier transform to the frequency domain, and is subjected to Gaussian high-pass filter, so as to filter out the backscattering light, and obtain the high-frequency part of the target signal light relative concentration; the minimum light intensity image I after the frequency domain Gaussian high-pass filter is obtained, that is:

[0017]

[0018] Wherein, and represent Fourier transform and inverse Fourier transform, respectively, H HP is the transfer function of the Gaussian high-pass filter.

[0019] Further, in step two, the acquisition method of the orthogonal polarization image pair by using the polarization degree correlation is: according to polarization degree maintaining and correlation calculation are performed to obtain the pretreated maximum light intensity image I , that is:

[0020]

[0021] It ensures that the polarization degree remains consistent before and after stretching and filtering.

[0022] Further, in the step three, the method for obtaining the polarization degree of backscattering light by processing the intensity image through a frequency domain low-pass filter and combining the optimal threshold factor is: first, the intensity image is transformed into the frequency domain through a Fourier function, then multiplied by the transfer function of the filter, and the result of the multiplication is returned to the spatial domain image through inverse Fourier transformation;

[0023]

[0024] In order to improve the image contrast, in the low contrast area, a threshold factor γ is introduced to optimize the accuracy of the estimated filter, the difference between the overall intensity B of the backscattering light and the original image intensity is defined as ΔI, and the threshold factor γ is set to improve the accuracy of the frequency domain low-pass filter:

[0025] ΔI(x,y) = |B(x,y)-I(x,y)|

[0026]

[0027] Wherein is the optimized backscattering light intensity estimation;

[0028] The DOP (polarization degree) estimation of backscattering light can be expressed as:

[0029]

[0030] Wherein is the DOP estimation of backscattering light, is the polarization degree estimation of target light.

[0031] Further, in the step three, the enhanced metric is used to evaluate the effect of despeckling to optimize the estimation of the polarization degree of the target light, and finally, the method for obtaining the image restoration combined with the physical model is: for the optimal estimation of P target , the enhanced metric (EME) is used to evaluate the effect of despeckling; at the same time, the optimal value of P target in [0, 1] with a step of 0.01 and the value of σ in [0, 0.3] with a step of 0.01 are searched as:

[0032]

[0033] Wherein, optional represents optimization, the image is divided into m×n regions with ordinal numbers (k1, k2), and are the maximum and minimum values of the image gray scale of each region respectively; here we divide the image into 80×80 regions to calculate the EME value, so the values of k1 and k2 are 1-80;

[0034] The final target signal light expression is obtained as:

[0035]

[0036] A clear recovery result was obtained.

[0037] (III) Beneficial Effects

[0038] Compared with existing technologies, this invention provides an underwater polarization imaging method based on frequency domain processing combined with global estimation, which has the following advantages:

[0039] This invention first preprocesses orthogonal polarization images by histogram stretching and high-pass filtering to reduce the influence of low-frequency scattered light, and obtains orthogonal polarization image pairs more accurately through polarization degree correlation. Subsequently, the intensity image is processed using a frequency domain low-pass filter, combined with an optimal threshold factor, to accurately estimate the polarization degree of backscattered light. Simultaneously, an enhancement metric (EME) is used to evaluate the descattering effect, thereby optimizing the estimation of the target light's polarization degree. Finally, image restoration is performed using a physical model. This method eliminates the need for human-computer interaction, effectively overcoming the shortcomings of traditional methods in restoring moving targets. Experimental studies were conducted on static and moving targets in turbid water, achieving satisfactory image restoration quality. Attached Figure Description

[0040] Figure 1 This is the overall flowchart of the present invention;

[0041] Figure 2 This is a model diagram of the polarization imaging system in the underwater environment that is the subject of this invention.

[0042] Figure 3 (a) is the total light intensity image; (b) and (c) are the images restored by the traditional method; (d) is the effect diagram of the actual application of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Example

[0045] like Figures 1-3 As shown, an embodiment of the present invention proposes an underwater polarization imaging method based on frequency domain processing combined with global estimation, comprising:

[0046] Step 1: Acquire multiple underwater images with different polarization directions, and establish target reflected light intensity model and backscattered light intensity model accordingly;

[0047] As Figure 2 shown is the polarization imaging system in the underwater environment of the present application, the collimating LED light source with the center wavelength of 625 nm is selected to realize the active illumination in the horizontal direction in combination with the linear polarizer, the starting light beam passes through the first polarizer 2 with the polarization direction set to the horizontal direction, the transparent water tank 5 and the turbid water 6 to irradiate on the target object 4 and the suspended particles 3, wherein the turbid water 6 is made by adding 20 ml of milk in the clear water, and the target object 4 is a plastic toy. The target object 4 reflects the starting light beam to obtain the target reflection light, and the suspended particles 3 scatter the starting light beam to obtain the backscattering light; the target reflection light and the backscattering light pass through the second polarizer 7 to irradiate on the CCD detector 8.

[0048] The polarization direction of the second polarizer 7 is rotated to 0° to obtain the polarization image I0; the polarization direction of the second polarizer 7 is rotated to 45° to obtain the polarization image I45; the polarization direction of the second polarizer 7 is rotated to 90° to obtain the polarization image I90; the polarization direction of the second polarizer 7 is rotated to 135° to obtain the polarization image I135. The Stokes vector is calculated by using the four polarization images:

[0049]

[0050] Wherein, I represents the total light intensity of the scene irradiation; Q is the intensity difference between the horizontal and vertical directions; U is the intensity difference between the 45° and 135° directions.

[0051] The overall degree of polarization is calculated:

[0052]

[0053] Wherein, P represents the overall degree of polarization.

[0054] Two orthogonal underwater polarization images are obtained

[0055]

[0056] Wherein, I || is the polarization image containing the maximum visible backscattering light intensity, which is recorded as the brightest image; I ⊥ is the polarization image containing the minimum visible backscattering light intensity, which is recorded as the darkest image.

[0057] I=T+B=I || +I ⊥

[0058] Wherein, I represents the total light intensity of the scene received by the detector, which can be decomposed in mutually orthogonal directions; similarly, T and B can also be decomposed; T represents the target reflected light, which is manifested as a clear image of the target in the underwater image, i.e. the image to be obtained; B represents the backscattered light, which is the main reason for the poor quality of the total light intensity I of the scene received by the detector.

[0059] The brightest image I || The darkest image I ⊥ Both can be composed of the target reflected light and the backscattered light. Specifically, the following relationship can be established:

[0060] I ⊥ = T ⊥ + B ⊥

[0061] I || = T || + B ||

[0062] Wherein, T || and B || are the brightest target reflected light intensity and the brightest backscattered light intensity, respectively, T ⊥ and B ⊥ are the darkest target reflected light intensity and the darkest backscattered light intensity, respectively.

[0063] The target reflected light polarization degree is calculated according to the brightest target reflected light intensity and the darkest target reflected light intensity, i.e.

[0064]

[0065] Wherein, P tar represents the target reflected light polarization degree.

[0066] The backscattered light polarization degree is calculated according to the brightest backscattered light intensity and the darkest backscattered light intensity, i.e.

[0067]

[0068] Wherein, P sca represents the backscattered light polarization degree.

[0069] According to the brightest image, the darkest image, the target reflected light polarization degree and the backscattered light polarization degree, a target reflected light intensity model is established:

[0070]

[0071] According to the brightest image, the darkest image, the target reflected light polarization degree and the backscattered light polarization degree, a backscattered light intensity model is established:

[0072]

[0073] Step two: First, the orthogonal polarization images are preprocessed by histogram stretching and high-pass filtering to reduce the influence of low-frequency scattered light, and orthogonal polarization image pairs are obtained more accurately through polarization degree correlation.

[0074] According to the classical polarization imaging model in underwater scattering environments, the total light intensity I(x,y) received by the camera during underwater imaging includes two parts: the target object signal light T(x,y) and the backscattered light B(x,y).

[0075] I(x,y)=T(x,y)+B(x,y)

[0076] =L(x,y)·t(x,y)+A ∞ [1-t(x,y)]

[0077] In the formula, L(x,y) represents the intensity of the unattenuated target signal light, i.e., the direct transmission light, t(x,y) represents the transmission coefficient of the transmission medium, and A ∞ This refers to backscattered light at an infinite distance.

[0078] Underwater polarization imaging technology is based on orthogonal polarization images I || (x,y) and I ⊥ The light intensity signal (x, y) is obtained and combined with the above physical model to restore a clear image of the target object. Here, I(x, y) represents the light intensity signal received by the camera without a linear polarizer.

[0079] I(x,y)=I || (x,y)+I ⊥ (x,y)

[0080] =[T || (x,y)+B || (x,y)]+[T ⊥ (x,y)+B ⊥ (x,y)]

[0081]

[0082] Where P(x,y) represents the degree of polarization.

[0083] For imaging in turbid media, the grayscale range of the image is compressed into a narrow band. This compression of the grayscale range can be mitigated by performing histogram stretching.

[0084]

[0085] It should be noted that a polarization relationship exists between two orthogonal polarized images. Therefore, independently stretching the grayscale range of the two orthogonal polarized images may disrupt this polarization relationship. To maintain the polarization relationship, we need to process one of the orthogonal images first using the histogram stretching method, and then process the other orthogonal image using DOP correlation.

[0086] In water, due to scattering and absorption, two orthogonally polarized images I are obtained. || (x,y) and I ⊥ Neither (x, y) accurately represents the light intensity and polarization information of the original target object. Using these two orthogonal polarization images for subsequent polarization image restoration will result in poor imaging quality. In reality, the reflected light from the target object and the backscattered light from the scattering medium differ in their spectral distributions within the polarization images. Relatively speaking, the backscattered light information from the scattering medium is mainly concentrated in the low-frequency components of the spectrum, while the target signal light information is mainly concentrated in the high-frequency components. Therefore, a frequency-domain high-pass filter can be designed to preprocess the stretched orthogonal polarization images, extracting the high-frequency components to separate the target signal light from the backscattered light.

[0087] Among them, due to the minimum light intensity diagram I perpendicular to the polarization direction of the active illumination line. ⊥ Scattered light in (x,y) is initially suppressed, making the target information more apparent. Therefore, I' after histogram stretching is chosen. ⊥ (x,y) undergoes frequency domain preprocessing. The stretched I' ⊥ The (x,y) signal is transformed to the frequency domain using Fourier transform and then subjected to a Gaussian high-pass filter to remove backscattered light, thus obtaining the relatively concentrated high-frequency portion of the target signal light. The corrected minimum intensity map is obtained after frequency-domain Gaussian high-pass filtering. for:

[0088]

[0089] in, and H represents the Fourier transform and the inverse Fourier transform, respectively. HP It is the transfer function of a Gaussian high-pass filter.

[0090] To maintain the original polarization relationship, the maximum light intensity diagram The same histogram stretching and high-pass filtering cannot be directly performed. Instead, based on... The maximum intensity map after preprocessing is obtained by performing polarization degree preservation and correlation calculations. :

[0091]

[0092] The function P(·) ensures that the degree of polarization remains consistent before and after stretching and filtering.

[0093] Orthogonal polarization image after frequency domain filtering preprocessing and It includes the high-frequency part of the original image spectrum, that is, it filters out a portion of the backscattered light, thus highlighting more information about the target object.

[0094] Considering the low-frequency characteristics of backscattered light, we perform a frequency-domain low-pass filter on the entire image to accurately estimate the overall intensity B, rather than selecting a background area without a target for calculation. The process involves first transforming the intensity image to the frequency domain using a Fourier function, then multiplying it by the filter's transfer function, and finally returning the result to the spatial domain image using an inverse Fourier transform.

[0095]

[0096] in and This represents the Fourier transform and inverse Fourier transform. The transfer function of a low-pass filter is:

[0097]

[0098] In the formula, D(u,v) is the distance from the data point to the center of the frequency domain, D0 is the cutoff frequency, and λ is the filter order. In this paper, we set the cutoff frequency to D0 = 50Hz and the filter order to λ = 2.

[0099] To further improve image contrast, especially in low-contrast regions, a threshold factor γ is introduced to optimize the accuracy of the estimation filter. The difference between the overall intensity B of the backscattered light and the intensity of the original image is defined as ΔI. By setting the threshold factor γ, the accuracy of the frequency domain low-pass filter is improved.

[0100] ΔI(x,y)=|B(x,y)-I(x,y)|

[0101]

[0102] in This is for the optimized backscatter intensity estimation.

[0103] The DOP estimate for backscattering can be expressed as:

[0104]

[0105] For P targetFor optimal estimation, we use the Enhancement Metric (EME) to evaluate the descattering effect. A higher EME value indicates better image quality. As a non-reference evaluation metric, EME can accurately and quantitatively assess image quality in most cases. Therefore, we simultaneously search for P values ​​within the range [0,1] with a step size of 0.01. target The optimal value and the value of σ with a step size of 0.01 within [0, 0.3] are:

[0106]

[0107] The image is divided into m×n regions using ordinal numbers (k1, k2). and These represent the maximum and minimum grayscale values ​​for each region. Here, we divide the image into 80×80 regions to calculate the EME values, so the values ​​of k1 and k2 range from 1 to 80.

[0108] Will Substituting the values, we obtain the final target signal light expression as follows:

[0109]

[0110] To verify the effectiveness of this invention, it was compared with images restored using two traditional underwater polarization imaging methods that require manual cropping of background areas without target regions to estimate parameters. Figure 3 (a) is the total light intensity image; the image is not very clear to the naked eye. For example... Figure 3 (b) and (c) are images restored using traditional methods. While these methods improve visibility somewhat, they do not achieve satisfactory results. This is because manually cropping background areas without target regions to estimate parameters lacks stability and reliability, and the evaluation metrics used in the algorithm's imaging process are singular, potentially leading to suboptimal results. In contrast, the method of this invention exhibits significantly higher visibility, such as... Figure 3 (d) indicates that the present invention estimates relevant parameters more accurately and can be processed automatically without human-computer interaction, making it a more efficient method.

[0111] To quantitatively evaluate image quality, in addition to contrast and sharpness, the Image Enhancement Evaluation Index (EME) was used to assess the quality of the restored image. A higher value indicates higher image quality. The results are shown in the table below:

[0112] Total light intensity map CLAHE Schechner The invention Contrast 7.7949 42.9413 95.9831 173.4208 Sharpness 15.0025 14.4855 20 26.1771 EME 2.0635 2.4719 3.1335 5.2084

[0113] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1.A method for underwater polarization imaging based on frequency domain processing combined with global estimation, characterized in that: The method comprises the following steps: Step one, obtaining multiple underwater images with different polarization directions, establishing an underwater polarization imaging model, and obtaining the polarization degree of target reflected light and backscattering light; Step two, performing histogram stretching and high-pass filter preprocessing on the orthogonal polarization images, and obtaining the orthogonal polarization image pair through polarization degree correlation; including: Total light intensity received by the camera when imaging underwater Containing target object signal light And backscattered light Two parts: Underwater polarization imaging technology is based on the acquisition of orthogonal polarization images and combined with a physical model to achieve a clear image of the target object, at which time is the light intensity signal received by the camera when no linear polarizer is placed: wherein is the degree of polarization; For imaging in turbid media, the gray level range of the image is compressed into a narrow band, and the compression of the gray level range is performed by executing the histogram stretching method to alleviate; By designing a frequency domain high-pass filter, the stretched orthogonal polarization images are preprocessed to extract high-frequency components, so as to separate the target signal light and the backscattering light; selecting the histogram-stretched performing a frequency domain pre-processing operation on the stretched performing a Fourier transform to the frequency domain and a Gaussian high-pass filtering to filter out the backscattered light, and obtaining a corrected minimum intensity map after the frequency domain Gaussian high-pass filtering is wherein and represent the Fourier transform and the inverse Fourier transform, respectively, is the transfer function of a Gaussian high-pass filter; According to Polarization degree retention and correlation calculation are performed to obtain a pretreated maximum light intensity map : where the function , which ensures that the degree of polarization remains consistent before and after stretching and filtering; Orthogonal polarized image pairs after frequency domain filtering pre-processing and contains high frequency parts in the original image spectrum; Step three, processing the intensity image through a frequency domain low-pass filter, combining an optimal threshold factor, estimating the polarization degree of the backscattering light, using an enhanced intensity metric to evaluate the effect of despeckling, optimizing the estimation of the polarization degree of the target light, and finally, combining a physical model to perform image restoration; including: First, the intensity image is transformed into the frequency domain through a Fourier function, then multiplied by the transfer function of the filter, and the result of the multiplication is returned to the spatial domain image through inverse Fourier transformation; wherein and denote the Fourier transform and the inverse Fourier transform, respectively, and the transfer function of the low-pass filter is H LP : In low contrast regions, a threshold factor is introduced to optimize the precision of the estimation filter, the overall intensity of backscattered light The difference to the original image intensity is defined as A threshold factor is set to improve the precision of the frequency domain low-pass filter ​ wherein is the optimized backscattered light intensity estimate; The polarization degree DOP estimation of the backscattering light is expressed as: wherein is an estimate of the degree of polarization DOP of the backscattered light, is an estimate of the degree of polarization of the target light; For the optimization estimation of the effect of de-scattering is evaluated using an enhanced metric EME while searching for the optimal value of in [0, 1] with step size 0.01 and the value of in [0, 0.3] with step size 0.01: wherein, optimal denotes the optimization, the image is divided into ordinal number of regions, and are the maximum and minimum image gray values of each region, respectively;​ By substituting the above equations into the equation (1), the final expression of the target signal light is obtained as follows: By substituting the above equations into the equation (1), the final expression of 。

Images