A polarization image defogging method and system based on target scene color recovery

Abstract

The application discloses a polarization image defogging method and system based on target scene color recovery, comprising: acquiring total light intensity graph and polarization differential graph, and calculating the atmospheric light value and spatial position at infinity according to the characteristics of low atmospheric light saturation and high brightness; using the partial polarization characteristics of the atmospheric light, extracting the atmospheric light polarization angle according to the spatial position of the atmospheric light at infinity, and combining the Stokes vector to calculate the atmospheric light polarization degree changing with the scene depth; based on the spatial correlation of the atmospheric scattering light edge jump, using the polarization degree and the chroma to construct a regularization constraint to correct the atmospheric scattering light; obtaining the restored image from the total light intensity graph, the atmospheric scattering light and the atmospheric light at infinity, and performing color deviation correction on the restored image pixel by pixel by using adaptive segmented equalization RGB color channel distribution; the application significantly improves the color fidelity and visibility of the target scene at different depths, improves the quality of the restored image, and has a significant effect on thick fog scenes.

Description

Technical Field

[0001] This invention relates to the fields of image processing and polarization vision technology, and more specifically to a method and system for dehazing polarized images based on target scene color restoration. Background Technology

[0002] Currently, existing image dehazing algorithms can be mainly divided into methods based on image enhancement, deep learning, and physical models. Image enhancement-based methods do not consider the nature of degradation, such as Retinex, wavelet transform, and multi-scale fusion, and have limitations in application scenarios. Deep learning-based methods still face many challenges due to the lack of publicly available large-scale real-world scene datasets. Physical model-based methods use priors and assumptions to estimate key parameters for restoring foggy images, such as dark channel priors, color attenuation priors, and nonlocal priors, but these methods are easily limited by the scene and have a limited scope of application.

[0003] In recent years, with the continuous development of polarization optical detection technology, polarization imaging can obtain more target and environmental information, enabling more effective target detection and identification.

[0004] Early algorithms were based on a polarization difference model, which uses polarization difference information to estimate atmospheric polarization by manually selecting sky regions to solve for atmospheric light and finally obtain a dehazed image. Later, based on this model, a method was proposed to use three or four images with polarization at different angles, combined with Stokes vectors, to estimate atmospheric light parameters and then invert the haze-free image.

[0005] Liang et al. automatically estimated the global parameter atmospheric polarization degree using polarization angle information. This method can avoid sky region selection, but color distortion occurs in some scenes. Shen et al. proposed a transmission map optimization scheme with depth-chroma compensation regularization and an image dehazing optimization scheme with chroma-depth compensation regularization, and performed denoising processing throughout the process. Shao et al. used atmospheric light gradient prior information as a constraint to separate the target layer and atmospheric layer information, and used atmospheric layer information to achieve polarization reconstruction. Zhang et al. used polarization optical dehazing method in the intensity channel of HIS color space for dehazing processing, and combined the principle of color constancy to correct image color distortion. Song et al. proposed a fog map reconstruction method based on orthogonal decomposition of atmospheric light polarization at infinity, and constructed a blind separation model of atmospheric scattering light with multiple regularization constraints to achieve accurate estimation of atmospheric scattering light, but this method has serious color distortion in areas with high fog concentration.

[0006] Therefore, how to overcome the problem of color distortion in restored images due to inaccurate estimation of atmospheric light parameters in dense fog environments is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a method and system for dehazing polarized images based on target scene color restoration to solve the technical problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A polarization image dehazing method based on target scene color restoration includes the following steps:

[0010] S1. Obtain the total light intensity map and polarization difference map of the original fog map, and calculate the atmospheric light value and spatial position at infinity based on the characteristics of low atmospheric light saturation and high brightness.

[0011] S2. Utilizing the partial polarization characteristics of atmospheric light, extract the polarization angle of atmospheric light based on the spatial position of atmospheric light at infinity, and combine it with the Stokes vector to calculate the degree of polarization of atmospheric light as it varies with scene depth.

[0012] S3. Based on the spatial correlation at the edge transition of atmospheric scattered light, regularization constraints are constructed using polarization degree and chromaticity to correct atmospheric scattered light;

[0013] S4. The restored image is obtained from the total light intensity map, the atmospheric scattered light, and the atmospheric light at infinity. The color deviation of the restored image is corrected pixel by pixel using an adaptive segmented equalization method for the RGB color channel distribution.

[0014] Preferably, step S1 includes the following:

[0015] S11. Perform minimum value filtering on the atmospheric polarization difference map, and then perform sorting on the filtered image to obtain the coordinate positions of the pixels with the highest brightness in the top 5%.

[0016] S12. Obtain the coordinates of the pixels with 5% saturation in the total light intensity map;

[0017] S13. Calculate the intersection coordinates of the two coordinate positions obtained in steps S11 and S12, which represent the spatial position with the highest probability of atmospheric light distribution at infinity;

[0018] S14. The atmospheric light value at infinity is obtained by averaging the pixel values ​​at the selected intersection coordinates.

[0019] Preferred, atmospheric light A at infinity ∞ The calculation method is as follows:

[0020]

[0021] in, Atmospheric polarization difference image A pThe coordinates of the pixels with the highest brightness in the image. The coordinates of the pixels at the bottom 5% of the saturation in the total light intensity I image are given.

[0022] Preferably, the specific content of step S2 is as follows:

[0023] S21. Following the frequency-first strategy, and combining the spatial positions of atmospheric light at infinity selected from the screening, find the corresponding value with the highest frequency of spatial position coordinates in the total polarization angle image, which is the polarization angle of atmospheric light.

[0024] S2. Calculate the atmospheric polarization degree as a function of scene depth, specifically:

[0025]

[0026] Among them, S I For information on the total atmospheric light intensity, S Q For atmospheric light polarization information, θ A This is the atmospheric light polarization angle.

[0027] Preferably, the specific content of step S3 is as follows:

[0028] S31. Atmospheric scattered light is calculated based on the atmospheric light polarization degree and combined with the atmospheric polarization difference image;

[0029] S32. Based on the high spatial correlation between atmospheric light polarization degree and scene depth, and taking advantage of the local smoothness of atmospheric light, a regularization objective function for atmospheric scattered light is constructed.

[0030] S33. Using the least squares method and the Fourier transform convolution theorem, and through iterative optimization, the objective function is solved to obtain the optimal atmospheric scattered light.

[0031] Preferably, the regularization objective function for the constructed atmospheric scattered light is as follows:

[0032]

[0033] The first term is the fidelity term, used to measure atmospheric light A before and after optimization. The fidelity between the two terms, the second term represents the regularization term of the weighted L1 norm, which is used to protect the edge details of atmospheric scattered light, λ is the regularization parameter used to control the smoothness of the image, A(x) and A(y) are the depths of local neighborhood pixels x and y in the atmospheric light image A, and W(x, y) is the weighting function, which adopts the joint function of atmospheric light polarization difference and chromaticity difference.

[0034]

[0035] Where c represents the RGB color channel, α is the weight of atmospheric light polarization and chromaticity, and δ is a very small constant. I represents the atmospheric light polarization degree of local neighboring pixels x and y in the atmospheric light image, respectively. c (x), I c (y) represents the light intensity values ​​of the original fog map of the local neighboring pixels x and y, respectively.

[0036] The preferred adaptive segmented equalization color channel formula is:

[0037]

[0038] Where c∈{r, g, b} represents the RGB color channels, J′ c The image after color cast correction, ρ c For adaptive color correction factor, This represents the average intensity of the RGB color channels; σ represents the average of the sum of the intensities of the three color channels; c σ is the standard deviation of the RGB color channel intensity. avg =(σ r +σ g +σ b ) / 3 represents the average of the sum of the standard deviations of the intensity of the three color channels.

[0039] A polarization image dehazing system based on target scene color restoration, based on the aforementioned polarization image dehazing method based on target scene color restoration, includes an atmospheric light estimation module at infinity, an atmospheric light polarization degree acquisition module, an atmospheric scattered light correction module, and a color shift correction module;

[0040] The atmospheric light estimation module at infinity is used to obtain the atmospheric light value and spatial location at infinity by combining the total light intensity map and polarization difference map based on the characteristics of low atmospheric light saturation and high brightness.

[0041] The atmospheric light polarization degree acquisition module is used to extract the atmospheric light polarization angle based on the spatial position of atmospheric light at infinity, utilizing the partial polarization characteristics of atmospheric light, and combining it with the Stokes vector to calculate the atmospheric light polarization degree that varies with scene depth.

[0042] The atmospheric scattered light correction module is used to correct atmospheric scattered light by constructing regularization constraints based on the spatial correlation at the edge transition of atmospheric scattered light and using polarization degree and chromaticity.

[0043] The color shift correction module is used to obtain a restored image from the atmospheric scattered light and atmospheric light at infinity. It adopts an adaptive segmented equalization method for the distribution of RGB color channels to perform color shift correction on the restored image pixel by pixel.

[0044] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for dehazing polarized images based on target scene color restoration.

[0045] A processing terminal includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the aforementioned polarization image dehazing method based on target scene color restoration.

[0046] As can be seen from the above technical solution, compared with the prior art, this invention discloses a polarization image dehazing method and system based on target scene color restoration. It estimates atmospheric light at infinity by combining the image characteristics of atmospheric light with the total intensity map and polarization difference map to reduce interference from white objects. Based on this, it extracts the polarization angle of atmospheric light using the partial polarization characteristics of atmospheric light at infinity, thereby automatically and accurately estimating the degree of polarization of atmospheric light, reducing estimation errors caused by manually selecting the sky region. Furthermore, based on the spatial correlation between the degree of polarization of atmospheric light and the chromaticity of fog-free images, it constructs an atmospheric light regularization constraint function, effectively improving the estimation accuracy of atmospheric light and enhancing the visibility and color fidelity of targets at different depths. In addition, it proposes an adaptive color equalization method to meet the task requirements of image restoration under different fog concentrations. This invention significantly improves the color fidelity and visibility of target scenes at different depths and further enhances the quality of the restored image, especially showing significant effects in image scenes with high fog and haze concentrations. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0048] Figure 1 The attached figure is a schematic diagram of a polarization image dehazing method based on target scene color restoration provided by the present invention;

[0049] Figure 2 The attached figure is a schematic diagram of the atmospheric light estimation distribution at infinity provided by the present invention;

[0050] Figure 3 The attached figure is a schematic diagram of the defogging results and color balance correction results in a dense fog scene provided by the present invention;

[0051] Figure 4 The attached figure is a schematic diagram of the restoration results of the target scene under different smog concentrations provided by the present invention;

[0052] Figure 5 The attached figure is a schematic diagram comparing the restoration results of a dense fog scene provided by the present invention with a clear, fog-free image.

[0053] Figure 6 The attached diagram is a comparative illustration of the defogging results of different methods for different haze concentrations in a target scene with a sky area provided by the present invention.

[0054] Figure 7 The attached diagram is a comparative illustration of the defogging results of different methods for different haze concentrations in a target scene with no sky, as provided by the present invention. Detailed Implementation

[0055] 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.

[0056] Example 1

[0057] This invention discloses a method for dehazing polarized images based on target scene color restoration, such as... Figure 1 As shown, it includes the following steps:

[0058] S1. Obtain the total light intensity map and polarization difference map of the original fog map, and calculate the atmospheric light value and spatial position at infinity based on the characteristics of low atmospheric light saturation and high brightness.

[0059] S2. Utilizing the partial polarization characteristics of atmospheric light, extract the polarization angle of atmospheric light based on the spatial position of atmospheric light at infinity, and combine it with the Stokes vector to calculate the degree of polarization of atmospheric light as it varies with scene depth.

[0060] S3. Based on the spatial correlation at the edge transition of atmospheric scattered light, regularization constraints are constructed using polarization degree and chromaticity to correct atmospheric scattered light;

[0061] S4. The restored image is obtained from the total light intensity map, the atmospheric scattered light, and the atmospheric light at infinity. The color deviation of the restored image is corrected pixel by pixel using an adaptive segmented equalization method for the RGB color channel distribution.

[0062] To further implement the above technical solution, the specific content of step S1 includes:

[0063] S11. Because the edges of bright white objects in the atmospheric polarization difference image may introduce errors into the estimation results, the atmospheric polarization difference image A... pMinimum filtering is performed, and then the filtered image is sorted in descending order to obtain the coordinates of the top 5% of pixels by brightness, denoted as .

[0064] S12. Obtain the coordinates of the pixels with 5% saturation in the total light intensity I image, and denote them as follows:

[0065] S13. Calculate the intersection coordinates of the two coordinate positions obtained in steps S11 and S12, representing the atmospheric light A at infinity. ∞ The spatial location with the highest probability distribution;

[0066] S14. Since atmospheric light at infinity is usually considered a global constant, the atmospheric light value at infinity is obtained by averaging the pixel values ​​at the selected intersection coordinate positions.

[0067] To further implement the above technical solution, atmospheric light A at infinity... ∞ The calculation method is as follows:

[0068]

[0069] in, Atmospheric polarization difference image A p The coordinates of the pixels with the highest brightness in the image. The coordinates of the pixels at the bottom 5% of the saturation in the total light intensity I image are given.

[0070] like Figure 2 The image shown is A without the influence of a bright white object. ∞ Estimated distribution map, Figure 2 In Figure (a), it can be seen that the pixel coordinates in the total light intensity I image include some white buildings; while Figure 2 In Figure (b), the filtered image effectively excludes white buildings. To further avoid estimation errors caused by the target scene in the image potentially having polarization characteristics, I and A are calculated. p The intersection coordinates in the diagram are used to reduce interference from white buildings, and the result is as follows: Figure 2 As shown in Figure (c), the spatial location with the highest probability corresponding to the intersection coordinates in the original figure is mostly in the sky region. White bright objects are well excluded, effectively reducing the influence of white objects and allowing for a more accurate estimation of atmospheric light values ​​at infinity.

[0071] To further implement the above technical solution, the specific content of step S2 is as follows:

[0072] S21. Following the frequency-first strategy, and combining the spatial position of atmospheric light at infinity selected from the screening, find the corresponding value with the highest frequency of spatial position coordinates in the total polarization angle image, which is the polarization angle of atmospheric light. This can effectively avoid interference from the fact that the target object in the image scene may also have some polarization characteristics, and also get rid of the limitations of the sky area and the bias coefficient.

[0073] S2. Calculate the atmospheric polarization degree as a function of scene depth, specifically:

[0074]

[0075] Among them, S I For information on the total atmospheric light intensity, S Q For atmospheric light polarization information, θ A This is the atmospheric light polarization angle.

[0076] To further implement the above technical solution, the specific content of step S3 is as follows:

[0077] S31. Atmospheric scattered light is calculated based on the atmospheric light polarization degree and combined with the atmospheric polarization difference image;

[0078] In this embodiment, the atmospheric light polarization degree is spatially highly correlated with the scene depth. If the depth changes of local neighboring pixels x and y in the atmospheric light A image are large, then the p-values ​​at these two pixel locations will be significantly different. A The deviation from A is usually also large, so the colors of the two pixels in the corresponding dehazed image will also have obvious differences. This situation often occurs at the edges of the target and sky regions in the image scene. Therefore, this correlation can be expressed as:

[0079] ||A(x)-A(y)||2∝||p A (x)-p A (y)||2∝||C(x)-C(y)||2

[0080] In the formula, C(x) and C(y) represent the colors of the restored image;

[0081] S32. In order to accurately estimate atmospheric scattered light so that the target scene in the image edge region of the restored result has high color fidelity, based on the high spatial correlation between atmospheric light polarization degree and scene depth, and utilizing the local smoothness of atmospheric light, a regularization objective function for atmospheric scattered light is constructed.

[0082] S33. Using the least squares method and the Fourier transform convolution theorem, and through iterative optimization, the objective function is solved to obtain the optimal atmospheric scattered light.

[0083] To further implement the above technical solution, the regularization objective function for atmospheric scattered light is specifically constructed as follows:

[0084]

[0085] The first term is the fidelity term, used to measure atmospheric light A before and after optimization. The fidelity between the two terms, the second term represents the regularization term of the weighted L1 norm, which is used to protect the edge details of atmospheric scattered light, λ mainly controls the smoothness of the image, the larger λ is, the smoother the image, A(x) and A(y) are the depths of local neighborhood pixels x and y in the atmospheric light image A, and W(x, y) is the weighting function, which adopts the joint function of atmospheric light polarization difference and chromaticity difference;

[0086]

[0087] Where c represents the RGB color channel, α is the weight of atmospheric light polarization and chromaticity, and δ is a very small constant to prevent this term from being 0. I represents the atmospheric light polarization degree of local neighboring pixels x and y in the atmospheric light image, respectively. c (x), I c (y) represents the light intensity values ​​of the original fog map of the local neighboring pixels x and y, respectively;

[0088] The optimal solution is obtained by minimizing the regularization objective function. When the polarization degree and color difference of adjacent pixels x and y change more, the weighting function is smaller, and the regularization constraint of atmospheric scattered light between pixels x and y is smaller, which has the effect of protecting edge jumps. Conversely, it will smooth the non-edge areas of atmospheric scattered light.

[0089] In practical applications, parameter λ is set to 0.2, parameter σ is set to 0.5, and parameter α is set to 0.6.

[0090] In this embodiment, the method for obtaining the restored image is as follows:

[0091]

[0092] Where A is the obtained atmospheric scattered light, A ∞ Let I be atmospheric light at infinity, and let I be the total luminous intensity.

[0093] Specifically:

[0094] According to the atmospheric scattering model, the total light intensity I of the image captured by the imaging detection system mainly consists of two parts: the direct transmitted light D from the target scene and the atmospheric scattered light A. The total light intensity reaching the imaging system can be expressed as:

[0095] I = D + A = Jt(x) + A ∞ (1-t(x))

[0096] In the formula, I represents a real foggy image, J represents a fog-free image, A∞ represents the atmospheric light value at infinity, and t(x) represents the transmittance. When the distribution of fog-suspended particles in the atmosphere is spatially uniform, the transmittance can be expressed as:

[0097] t(x)=e -βd(x)

[0098] In the formula, d(x) is the scene depth, and β is the atmospheric scattering coefficient;

[0099] Assuming that the direct transmitted light D, which suffers from scene attenuation, has negligible polarization characteristics on a macroscopic scale, while the atmospheric scattered light A becomes partially polarized after being scattered by suspended particles in the atmosphere, we can obtain:

[0100] I p =Ip=I ⊥ -I || ≈A p

[0101] In the formula, I ⊥ I || These represent the maximum and minimum light intensity maps, I. p For the total polarization difference plot, A p This is an atmospheric polarization difference image, represented by the polarization portion of atmospheric scattered light;

[0102] According to the definition of polarization degree, the total polarization degree p and the atmospheric optical polarization degree p A The expression is:

[0103]

[0104]

[0105] In the formula, A ⊥ A || These represent the maximum and minimum atmospheric light intensity diagrams, respectively.

[0106] Finally, based on the above parameters, the expression for a fog-free image can be derived as follows:

[0107]

[0108] To further implement the above technical solution, the adaptive segmented equalization color channel formula is as follows:

[0109]

[0110] Where c∈{r, g, b} represents the RGB color channels, J′ c The image after color cast correction, ρ c For adaptive color correction factor, This represents the average intensity of the RGB color channels; σ represents the average of the sum of the intensities of the three color channels; c σ is the standard deviation of the RGB color channel intensity. avg =(σ r +σ g +σ b ) / 3 represents the average of the sum of the standard deviations of the intensity of the three color channels.

[0111] In this embodiment, to accurately assess whether color distortion exists in the restoration result of the dehazing algorithm presented in this paper, a color cast factor K is introduced as an evaluation index to measure the degree of color distortion in the dehazed image. Experimental statistical analysis revealed that when the color cast factor K ≥ 0.485, the target scene is prone to significant color distortion, and the larger the K value, the more severe the distortion. In this case, an adaptive color equalization method using the adaptive piecewise equalization color channel formula must be employed to adaptively correct the color cast, thereby restoring the basic color information of the target object and the sky scene. The restoration result is as follows: Figure 3 As shown, therefore, combining the adaptive color equalization method can meet the restoration requirements of the dehazing algorithm of this invention in dense fog scenes, making the colors of the dehazed image more consistent with the visual effect of the human eye.

[0112] In practical applications, experiments and result analysis were conducted on a polarization image dehazing method based on target scene color restoration according to the present invention. Since there is currently a lack of standard real-world foggy scene polarization image datasets to evaluate the algorithm's performance, an atmospheric polarization information observation platform was built to acquire a large amount of real and effective foggy polarization image data. The platform's acquisition equipment consisted of a color polarization camera (model SZ-FGEB500C) and related atmospheric environment monitoring equipment (such as visibility meters and PM2.5 monitoring sensors). The experimental data consisted of color polarization image data of different haze concentrations collected by the observation platform under real foggy scenes. By controlling the platform, three original polarized foggy images (I0, I10, I20, I30, I40, I50, I60, I70, I80, I90, I10, I10, I2 ... 60 and I 120 The image resolution is 1910×1190.

[0113] To classify the collected data, atmospheric visibility parameters and the Fog Perception Evaluator (FADE) index were introduced to group the experimental data. Based on the FADE values ​​of the fog images and the statistical characteristics of scene visibility information, the experimental dataset was divided into light fog (0 < FADE < 5.7, 1km < visibility < 3km), moderate fog (5.7 < FADE < 7.9, 200m < visibility < 1km), and dense fog (7.9 < FADE < 9.0, visibility < 200m).

[0114] Example 2

[0115] This embodiment compares the robustness and effectiveness of the polarization image dehazing method based on target scene color restoration proposed in this invention under different fog concentration scenarios. Image data was selected from datasets of light fog, moderate fog, and dense fog scenarios for restoration. The original image is shown below. Figure 4 As shown in (a), (c), and (e), the restoration results are as follows: Figure 4 As shown in (b), (d), and (f), the results demonstrate that the present invention achieves good restoration effects for target scenes with different smog concentrations, realizes clear visibility of near and far targets, restores the true color information of target scenes at different depths, and has high color fidelity for targets such as white buildings, lakes, and bushes in different scenes in the image; at the same time, no obvious color distortion or halo effect appears at the edges of the target and sky areas in the restored image.

[0116] Example 3

[0117] This embodiment further explores the performance of the proposed method in restoring the color fidelity of the target scene. It introduces a set of color feature indices to analyze the differences in color channel histogram distribution and feature statistics between the restored image in a foggy scene and a fog-free image on a clear day. Furthermore, it utilizes the color cast factor K to measure the degree of color cast in the dehazed image. The color feature indices include the mean difference coefficient μ. d (A larger value indicates a greater difference in the mean between RGB channels), standard deviation coefficient of difference σ d (A larger value indicates a greater difference in standard deviation between RGB channels), and histogram cosine similarity HS (a larger value indicates a greater relative degree of histogram distribution between RGB channels). Experimental analysis results are as follows: Figure 5 As shown, comparison Figure 5 Results (d), (e), and (f) show that the color channel histogram distribution of the restored image from this invention covers grayscale values ​​within the range of [0, 255]. The increase in histogram similarity (HS) and decrease in the color cast factor (K) indicate that the overall color distortion of the image is relatively small, and the histogram distribution shift of the three color channels is small, consistent with a natural image on a sunny day (e.g., ...). Figure 5 The distribution pattern of the color channel histogram in (c) of the image; and, according to Figure 5 The mean difference coefficient μ of the three images (a), (b), and (c) in the figure. d and the coefficient of variation of standard deviation σ d The values ​​were analyzed, and it was found that the differences in the mean and standard deviation of the three color channels in the restored result were reduced, and they were closer to the color feature statistics of the sunny image. This indicates that the color distribution of the processed image is more concentrated, the statistical feature differences between the three channels are reduced, and the distribution characteristics of a natural and undistorted image are met.

[0118] Through experimental analysis of polarized images of foggy days under different concentrations in Examples 2 and 3, it can be seen that the present invention has a good defogging effect on light fog, medium fog and dense fog scene data, effectively reduces the overall image color distortion, and subjectively is closer to the visual effect of natural and undistorted images. Therefore, the present invention can adapt to the defogging task of different fog and haze concentration scenes, and has significant advantages, especially for the restoration of target color fidelity.

[0119] Example 4

[0120] This embodiment compares the restoration results of the proposed method with those of classic dehazing algorithms such as BCCR, DCP, GPLPF, and PLF. The selected comparative experimental dataset contains real foggy images under different haze concentrations, mainly divided into target scenes with no sky and with sky, to explore the impact of whether or not a sky region is present on the restoration results.

[0121] Specifically: The total light intensity image obtained from the Stokes vector is used as input to BCCR and DCP, while the original foggy polarization image is used as input to GPLPF, PLF, and the method of this invention. The results of the four comparison methods and the restoration of foggy images in scenes with sky are as follows: Figure 6 As shown, and the image restoration results in a foggy scene without a sky, as shown in the figure. Figure 7 As shown, the original image's haze concentration is divided into light fog (… Figure 6 L1 and L2 in the middle, Figure 7 L3 and L4 in the middle, and medium fog ( Figure 6 M1 and M2 in the middle, Figure 7 M3 and M4), dense fog ( Figure 6 D1 and D2 in the middle, Figure 7 As shown in groups D3 and D4, this invention, along with the four comparative methods, demonstrates good dehazing effects in lightly foggy scenes. The BCCR and DCP methods result in overall darker brightness and noticeable color casts on target objects. The GPLPF and PLF methods restore near-field targets more naturally and exhibit high overall brightness and color fidelity in scenes without sky areas, but show slight halo effects in scenes with sky areas. However, in moderate and dense fog scenes, this invention produces more natural dehazing results than BCCR, GPLPF, and PLF. Furthermore, the proposed method does not exhibit significant halo effects or color distortion in dense fog or sky areas. It also demonstrates good dehazing effects on distant objects, resulting in more detailed distant objects with less noise.

[0122] In this embodiment, in order to objectively evaluate the image quality after dehazing, the natural image quality assessment index NIQE, the fog density assessment index FADE, the blind image spatial domain assessment index BRISQUE, the color cast factor index K, and the color channel histogram distribution difference index CHE are used to quantitatively analyze the comparative experimental data. Among them, the smaller the NIQE, FADE, and BRISQUE values, the better the image quality and the clearer the image. The color cast factor K and the color channel histogram distribution difference index CHE are used to measure the degree of color cast of the image. The larger the value, the greater the degree of color distortion of the image.

[0123] Tables 1-3 present the quantitative results of the present invention and four comparative defogging methods for NIQE, FADE, BRISQUE, K, and CHE in different haze concentration scenarios:

[0124] Table 1 compares the performance metrics of the algorithm presented in this paper with those of different defogging algorithms in a light fog scenario:

[0125]

[0126] Table 2 compares the performance indicators of the algorithm presented in this paper with those of different defogging algorithms in a moderate fog scenario:

[0127]

[0128] Table 3 shows the performance indicators of the algorithm presented in this paper and different defogging algorithms in dense fog scenarios:

[0129]

[0130]

[0131] It can be seen that the FADE value of the present invention is not as good as that of the DCP method in light fog and medium fog scenes, but the difference is small in numerical terms; however, in dense fog scenes, the fog concentration of the image obtained by the present invention is lower than that of the DCP method. Obviously, the present invention is superior to other methods in terms of NIQE, BRISQUE, K and CHE, and compared with the second-best result, the calculated mean values ​​of these four indicators are improved by 10.89%, 14.29%, 40.13% and 35.80%, respectively. The indicator analysis results show that the method proposed in the present invention has the best defogging effect, the restored image is closest to the natural image, and the color reproduction of the target scene is the highest.

[0132] A polarization image dehazing system based on target scene color restoration, which is based on a polarization image dehazing method based on target scene color restoration, includes an atmospheric light estimation module at infinity, an atmospheric light polarization degree acquisition module, an atmospheric scattered light correction module, and a color shift correction module.

[0133] The atmospheric light estimation module at infinity is used to obtain the atmospheric light value and spatial location at infinity by combining the total light intensity map and polarization difference map based on the characteristics of low atmospheric light saturation and high brightness.

[0134] The atmospheric light polarization degree acquisition module is used to extract the atmospheric light polarization angle based on the spatial position of atmospheric light at infinity, utilizing the partial polarization characteristics of atmospheric light, and combining it with the Stokes vector to calculate the atmospheric light polarization degree that varies with scene depth.

[0135] The atmospheric scattered light correction module is used to correct atmospheric scattered light by constructing regularization constraints based on the spatial correlation at the edge transition of atmospheric scattered light and using polarization degree and chromaticity.

[0136] The color shift correction module is used to obtain a restored image from the atmospheric scattered light and atmospheric light at infinity. It adopts an adaptive segmented equalization method for the distribution of RGB color channels to perform color shift correction on the restored image pixel by pixel.

[0137] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a polarization image dehazing method based on target scene color restoration.

[0138] A processing terminal includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements a polarization image dehazing method based on target scene color restoration.

[0139] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0140] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for dehazing polarized images based on target scene color restoration, characterized in that, Includes the following steps: S1. Obtain the total light intensity map and polarization difference map of the original fog map, and calculate the atmospheric light value and spatial position at infinity based on the characteristics of low atmospheric light saturation and high brightness. S2. Utilizing the partial polarization characteristics of atmospheric light, extract the polarization angle of atmospheric light based on the spatial position of atmospheric light at infinity, and combine it with the Stokes vector to calculate the degree of polarization of atmospheric light as it changes with scene depth. S3. Based on the spatial correlation at the edge transition of atmospheric scattered light, regularization constraints are constructed using polarization degree and chromaticity to correct atmospheric scattered light; S4. The restored image is obtained from the total light intensity map, the atmospheric scattered light, and the atmospheric light at infinity. The color deviation of the restored image is corrected pixel by pixel using the adaptive segmented equalization method of RGB color channel distribution. The formula for adaptive segmented equalization of color channels is: in, Indicates the RGB color channel. This is the image after color cast correction. For adaptive color correction factor, , Atmospheric scattered light Atmospheric polarization degree of local neighborhood pixels x and y in the image; This represents the average intensity of the RGB color channels; , representing the average of the sum of the intensities of the three color channels; The standard deviation of the RGB color channel intensity; , which represents the average of the sum of the standard deviations of the intensity of the three color channels.

2. The polarization image dehazing method based on target scene color restoration according to claim 1, characterized in that, The specific content of step S1 includes: S11. Perform minimum value filtering on the atmospheric polarization difference map, and then perform sorting on the filtered image to obtain the coordinates of the top 5% of the pixels in terms of brightness. S12. Obtain the coordinates of the pixels with 5% saturation in the total light intensity map; S13. Calculate the intersection coordinates of the two coordinate positions obtained in steps S11 and S12, which represent the spatial position with the highest probability of atmospheric light distribution at infinity; S14. The atmospheric light value at infinity is obtained by averaging the pixel values ​​at the selected intersection coordinates.

3. The polarization image dehazing method based on target scene color restoration according to claim 1, characterized in that, Atmospheric light at infinity The calculation method is as follows: in, Atmospheric polarization difference image The coordinates of the pixels with the highest brightness in the image. Total light intensity The coordinates of the pixels with 5% saturation in the image.

4. The polarization image dehazing method based on target scene color restoration according to claim 1, characterized in that, The specific content of step S2 is as follows: S21. Following the frequency-first strategy, and combining the spatial positions of atmospheric light at infinity selected from the screening, find the corresponding value with the highest frequency of spatial position coordinates in the total polarization angle image, which is the polarization angle of atmospheric light. S2. Calculate the atmospheric polarization degree as a function of scene depth, specifically: in, This represents the total atmospheric light intensity information. Information on atmospheric light polarization. This is the atmospheric light polarization angle.

5. The polarization image dehazing method based on target scene color restoration according to claim 1, characterized in that, The specific content of step S3 is as follows: S31. Atmospheric scattered light is calculated based on the atmospheric light polarization degree and combined with the atmospheric polarization difference image; S32. Based on the high spatial correlation between atmospheric light polarization degree and scene depth, and taking advantage of the local smoothness of atmospheric light, a regularization objective function for atmospheric scattered light is constructed. S33. Using the least squares method and the Fourier transform convolution theorem, and through iterative optimization, the objective function is solved to obtain the optimal atmospheric scattered light.

6. The polarization image dehazing method based on target scene color restoration according to claim 5, characterized in that, The regularization objective function for atmospheric scattered light is as follows: The first item is the fidelity item, which measures the atmospheric light before and after optimization. and The fidelity between them, the second item represents the weighted average. The norm regularization term is used to preserve edge details of atmospheric scattered light; λ is a regularization parameter used to control the smoothness of the image; A(x) and A(y) are atmospheric light images. The depth of the local neighborhood pixels x and y. The weighting function is a joint function of atmospheric light polarization difference and chromaticity difference; in, Indicates the RGB color channel. It is the weight of atmospheric light polarization degree and chromaticity. It is a very small constant. , These represent the atmospheric light polarization degrees of local neighboring pixels x and y in the atmospheric light image, respectively. , These are the light intensity values ​​of the original fog map for local neighboring pixels x and y, respectively.

7. A polarization image dehazing system based on target scene color restoration, characterized in that, A polarization image dehazing method based on target scene color restoration according to any one of claims 1-6 includes an atmospheric light estimation module at infinity, an atmospheric light polarization degree acquisition module, an atmospheric scattered light correction module, and a color deviation correction module. The atmospheric light estimation module at infinity is used to obtain the atmospheric light value and spatial location at infinity by combining the total light intensity map and polarization difference map based on the characteristics of low atmospheric light saturation and high brightness. The atmospheric light polarization degree acquisition module is used to extract the atmospheric light polarization angle based on the spatial position of atmospheric light at infinity, utilizing the partial polarization characteristics of atmospheric light, and combining it with the Stokes vector to calculate the atmospheric light polarization degree that varies with scene depth. The atmospheric scattered light correction module is used to correct atmospheric scattered light by constructing regularization constraints based on the spatial correlation at the edge transition of atmospheric scattered light and using polarization degree and chromaticity. The color shift correction module is used to obtain a restored image from the atmospheric scattered light and atmospheric light at infinity. It adopts an adaptive segmented equalization method for the distribution of RGB color channels to perform color shift correction on the restored image pixel by pixel.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements a polarization image dehazing method based on target scene color restoration as described in any one of claims 1-6.

9. A processing terminal, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, characterized in that, When the processor executes a computer program, it implements a polarization image dehazing method based on target scene color restoration as described in any one of claims 1-6.

Images