Image Dehazing Method and Device Based on Polarization Characteristics and Atmospheric Transmission Model

Abstract

The present application relates to an image defogging method and apparatus based on polarization characteristics and an atmospheric transmission model. The method includes: calculating a scene polarization degree image, a scene polarization angle image, and a scene light intensity image based on an initial polarization image; performing joint segmentation on the scene polarization angle image and the scene light intensity image, and solving to obtain an atmospheric polarization degree and an atmospheric light intensity at infinity; calculating a polarization difference image based on the scene polarization degree image and the scene light intensity image; deriving a target polarization degree image using a Gaussian blur algorithm; calculating an initial transmittance image based on the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image, and the target polarization degree image, and optimizing it through a sky region and a guided filtering algorithm to obtain a final transmittance image, thereby reconstructing a fog-free image. Using this method, it is possible to achieve image reconstruction under different scenes and different foggy weather conditions, and effectively restore image details and reduce image noise.

Description

Technical Field

[0001] This application relates to the fields of optics and image processing, and particularly to an image defogging method and device based on polarization characteristics and an atmospheric transmission model. Background Art

[0002] Under foggy conditions, there is sufficient and stable water vapor in the atmosphere, and a large number of suspended particles formed by tiny water droplets are suspended. Due to the strong scattering characteristics of the suspended particles themselves, light is scattered and attenuated during propagation, resulting in a reduction in visibility. When imaging a target under foggy conditions, the resulting image usually has characteristics such as low visibility, low contrast, blurred detail information, and even color distortion, which cannot meet the requirements for image quality in fields such as target detection and recognition. Therefore, in order to better improve the accuracy and reliability of target detection and recognition, and improve the detail quality of foggy images, it is an extremely urgent practical need to study methods for clear imaging under foggy conditions.

[0003] Currently, common defogging algorithms are mainly divided into two categories: image enhancement-based defogging and atmospheric model-based defogging. The main idea of image enhancement-based defogging is to improve the imaging effect of the target from the perspective of enhancing image contrast. Currently, common image enhancement algorithms can be divided into two categories from the processing scope: global enhancement and local enhancement. Representative algorithms for global image enhancement include the Retinex algorithm, etc. The Retinex algorithm is an algorithm used to describe color constancy. It clarifies that the color of an object depends on its reflection ability for light of each band, and has nothing to do with the absolute value of the reflection intensity, laying a solid foundation for image enhancement theory. However, the traditional Retinex algorithm is susceptible to the influence of light intensity values when processing over-bright images or when halos appear in local regions of the image. The representative of the local enhancement algorithm is the local variance algorithm, which calculates the variance of local regions of the image to determine the degree of image enhancement, and stretches the local gray level to achieve the effect of image enhancement, but the defogging effect is not robust.

[0004] The image defogging technology based on the atmospheric transmission model starts from the physical mechanism of the degradation of foggy images, constructs a model for foggy imaging, and uses the physical model to invert the imaging structure in the fog-free state. This defogging technology starts from the essence of image degradation to solve the problems under foggy weather conditions, such as the image degradation caused by the scattering and transmission attenuation of light by suspended particles in the atmosphere. A representative algorithm is the dark channel prior defogging algorithm. By analyzing a large amount of experimental data under fog-free weather, it is found that at least one channel of the pixel values in the three channels of the pixel points in the non-sky area is extremely low and infinitely close to zero. According to this prior knowledge and the atmospheric transmission model, the transmittance is estimated, and finally the fog-free image is inverted. The dark channel defogging algorithm has the characteristics of stability and simple and easy-to-understand principle. However, the dark channel defogging algorithm is prone to distortion problems when processing strong light images, and is prone to color deviation and block effect when processing images with uneven color distribution. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide an image defogging method and device based on polarization characteristics and atmospheric transmission model that can realize the reconstruction of fog-free images under different scenarios and different foggy weather conditions.

[0006] An image defogging method based on polarization characteristics and atmospheric transmission model, the method includes:

[0007] According to the initial polarization images at four angles collected by a four-channel optical polarization detector, calculate the Stokes vector according to the initial polarization images, and calculate the scene polarization degree image, the scene polarization angle image and the scene light intensity image according to the Stokes vector;

[0008] Perform joint segmentation on the area to be segmented according to the gradient information and light intensity information in the scene polarization angle image and the scene light intensity image to obtain the sky area, calculate the atmospheric polarization degree and the atmospheric light intensity at infinity according to the mask of the sky area, the scene polarization degree image and the scene light intensity image; calculate the polarization difference image according to the scene polarization degree image and the scene light intensity image;

[0009] Use the Gaussian blur algorithm to filter the initial polarization images to obtain the atmospheric polarization images, calculate the target polarization images according to the atmospheric polarization images, and calculate the target polarization degree images according to the target polarization images;

[0010] Calculate the initial transmittance image according to the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image and the target polarization degree image, determine the sky area of the initial transmittance image through the mask of the sky area, optimize and correct the initial transmittance image according to the sky area of the initial transmittance image and the guided filter algorithm to obtain the final transmittance image, and reconstruct the fog-free image according to the final transmittance image.

[0011] In one embodiment, calculating a Stokes vector from an initial polarization image includes:

[0012] Calculating a Stokes vector from an initial polarization image, expressed as

[0013] S = [S0, S1, S2, S3] T

[0014]

[0015] where S0 represents the total light intensity received by a four-channel optical polarization detector, S1 represents the intensity difference between the horizontal component light intensity and the vertical component light intensity, S2 represents the intensity difference between the 45° linear polarization and the 135° linear polarization, S3 represents the intensity difference between the right-handed circular polarization and the left-handed circular polarization, and I(0), I(45), I(90), and I(135) represent the initial polarization images at four angles.

[0016] In one embodiment, jointly segmenting a region to be segmented based on gradient information and light intensity information in a scene polarization angle image and a scene light intensity image to obtain a sky region includes:

[0017] Extracting the gradient information of the region to be segmented in the scene polarization angle image and the scene light intensity image, and setting a first threshold and a second threshold to judge the change of the gradient information;

[0018] Extracting the light intensity information of the region to be segmented in the scene light intensity image, and setting a third threshold to judge the change of the light intensity information,

[0019] Jointly segmenting the region to be segmented according to the first threshold, the second threshold, and the third threshold to obtain a sky region.

[0020] In one embodiment, jointly segmenting the region to be segmented according to the first threshold, the second threshold, and the third threshold to obtain a sky region includes:

[0021] When the gradient information of the region to be segmented in the scene polarization angle image is lower than the first threshold, the gradient information of the region to be segmented in the scene light intensity image is lower than the second threshold, and the light intensity information of the region to be segmented in the scene light intensity image is greater than the third threshold, it is judged that the region to be segmented is a sky region; otherwise, it is judged that the region to be segmented is a target region;

[0022] Assigning the pixel points in the sky region to 1 according to a binarization operation, and assigning the pixel points in the target region to 0. The sky region is expressed as

[0023] Sky(x, y) = 1((G AOP (x, y) < th1) and (G I(x,y) < th2) and (I(x,y) > th3)

[0024] Among them, G AOP G(x,y) represents the gradient information of the region to be segmented in the scene polarization angle image, and G I (x,y) represents the gradient information of the region to be segmented in the scene light intensity image, I(x,y) represents the light intensity information of the region to be segmented in the scene light intensity image, th1, th2, and th3 respectively represent the first threshold, the second threshold, and the third threshold, AOP represents the scene polarization angle image, and I represents the scene light intensity image.

[0025] In one embodiment, calculations are performed based on the mask of the sky region, the scene polarization degree image, and the scene light intensity image to obtain the atmospheric polarization degree and the atmospheric light intensity at infinity, including:

[0026] Multiply the mask of the sky region by the scene polarization degree image to obtain the sky region scene polarization degree image. By statistically analyzing the polarization degrees of all pixel points in the sky region scene polarization degree image, select the polarization degree that appears most frequently as the atmospheric polarization degree;

[0027] Multiply the mask of the sky region by the scene light intensity image to obtain the sky region scene light intensity image. Sort the intensity values of all pixel points in the sky region scene light intensity image in descending order, and select the average value of the intensity values of the top one hundred pixel points as the atmospheric light intensity at infinity.

[0028] In one embodiment, the Gaussian blur algorithm is used to filter the initial polarization image to obtain the atmospheric polarization image, including:

[0029] Use the Gaussian blur algorithm to filter the initial polarization images at four angles to obtain the atmospheric polarization images at four angles. The filtering formula is expressed as

[0030] A = max(min(aB, I′), 0)

[0031] B = Gauss(I′) - Gauss(|I′ - Gauss(I′)|)

[0032] Among them, I′ represents the image to be filtered, B represents the intermediate filtering value after initial filtering, A represents the finally obtained atmospheric light intensity, and a represents the adjustment parameter.

[0033] In one embodiment, calculations are performed based on the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image, and the target polarization degree image to obtain the initial transmittance image, which is expressed as:

[0034]

[0035] Among them, P d represents the target degree of polarization image, I d represents the polarization difference image, P a represents the atmospheric degree of polarization, A inf represents the atmospheric light intensity at infinity.

[0036] In one embodiment, the sky region of the initial transmittance image is determined by a mask of the sky region, and the initial transmittance image is optimized and corrected according to the sky region of the initial transmittance image and the guided filtering algorithm to obtain the final transmittance image, including:

[0037] The sky region of the initial transmittance image is determined by a mask of the sky region. When the proportion of the sky region of the initial transmittance image exceeds half of the initial transmittance image, the initial transmittance image is optimized according to the sky region of the initial transmittance image to obtain a preliminary optimized transmittance image, and the preliminary optimized transmittance image is corrected according to the guided filtering algorithm to obtain the final transmittance image; when the proportion of the sky region of the initial transmittance image does not exceed half of the initial transmittance image, the initial transmittance image is optimized according to the guided filtering algorithm to obtain the final transmittance image.

[0038] In one embodiment, when the proportion of the sky region of the initial transmittance image exceeds half of the initial transmittance image, the initial transmittance image is optimized according to the sky region of the initial transmittance image to obtain a preliminary optimized transmittance image, including:

[0039] When the proportion of the sky region of the initial transmittance image exceeds half of the initial transmittance image, random assignment processing is performed on the sky region of the initial transmittance image to obtain a preliminary optimized transmittance image with the transmittance maintained within the range of 0.05±0.01.

[0040] An image defogging device based on polarization characteristics and an atmospheric transmission model, the device includes:

[0041] An initial polarization image preliminary processing module, configured to calculate the Stokes vector according to four initial polarization images collected by a four-channel optical polarization detector, and calculate the scene polarization degree image, the scene polarization angle image, and the scene light intensity image according to the Stokes vector;

[0042] A sky region joint segmentation module, configured to perform joint segmentation on the region to be segmented according to the gradient information and light intensity information in the scene polarization angle image and the scene light intensity image to obtain the sky region, calculate the atmospheric polarization degree and the atmospheric light intensity at infinity according to the mask of the sky region, the scene polarization degree image, and the scene light intensity image; calculate the polarization difference image according to the scene polarization degree image and the scene light intensity image;

[0043] The target polarization degree image acquisition module is used to perform filtering processing on the initial polarization image by using the Gaussian blur algorithm to obtain the atmospheric polarization image, calculate the target polarization image based on the atmospheric polarization image, and calculate the target polarization degree image based on the target polarization image;

[0044] The haze-free image reconstruction module is used to calculate based on the scene light intensity image, atmospheric polarization degree, atmospheric light intensity at infinity, polarization difference image, and target polarization degree image to obtain the initial transmittance image, determine the sky region of the initial transmittance image through the mask of the sky region, optimize and correct the initial transmittance image according to the sky region of the initial transmittance image and the guided filter algorithm to obtain the final transmittance image, and reconstruct the haze-free image according to the final transmittance image.

[0045] The above image dehazing method and device based on polarization characteristics and atmospheric transmission model first calculate the scene polarization degree image, scene polarization angle image, and scene light intensity image according to the Stokes vectors corresponding to the initial polarization images at four angles; perform joint segmentation on the region to be segmented according to the gradient information and light intensity information in the scene polarization angle image and scene light intensity image, and solve the atmospheric polarization degree and atmospheric light intensity at infinity according to the segmented sky region; calculate according to the scene polarization degree image and scene light intensity image to obtain the polarization difference image; then use the Gaussian blur algorithm to calculate the atmospheric polarization images at each angle, and further deduce the target polarization degree image; then calculate the initial transmittance image according to the obtained scene light intensity image, atmospheric polarization degree, atmospheric light intensity at infinity, polarization difference image, and target polarization degree image, determine the sky region of the initial transmittance image through the mask of the sky region, and then optimize and correct the initial transmittance image according to the sky region of the initial transmittance image and the guided filter algorithm to obtain the final transmittance image, and finally reconstruct the haze-free image according to the final transmittance image. Using this method can realize the reconstruction of images under different scenes and different foggy weather conditions, can effectively restore image details, reduce image noise, and has good algorithm universality. Description of the Drawings

[0046] Figure 1 It is a schematic flowchart of the image dehazing method based on polarization characteristics and atmospheric transmission model in an embodiment;

[0047] Figure 2 It is the initial polarization images at four angles in an embodiment: Figure 2 (a) is the initial polarization image at 0°, Figure 2 (b) is the initial polarization image at 45°, Figure 2 (c) is the initial polarization image at 90°, Figure 2 (d) is the initial polarization image at 135°;

[0048] Figure 3 The degree-of-polarization image and the polarization-angle image of a scene in an embodiment: Figure 3 (a) is the degree-of-polarization image of the scene, Figure 3 (b) is the polarization-angle image of the scene;

[0049] Figure 4 The schematic diagram of the statistical result of the degree of polarization of the atmosphere in an embodiment;

[0050] Figure 5 The polarization images of the atmosphere at four angles in an embodiment: Figure 5 (a) is the polarization image of the atmosphere at 0°, Figure 5 (b) is the polarization image of the atmosphere at 45°, Figure 5 (c) is the polarization image of the atmosphere at 90°, Figure 5 (d) is the polarization image of the atmosphere at 135°;

[0051] Figure 6 The polarization images of the target at four angles in an embodiment: Figure 6 (a) is the polarization image of the target at 0°, Figure 6 (b) is the polarization image of the target at 45°, Figure 6 (c) is the polarization image of the target at 90°, Figure 6 (d) is the polarization image of the target at 135°;

[0052] Figure 7 The degree-of-polarization image of the target in an embodiment;

[0053] Figure 8 The initial transmittance image in an embodiment;

[0054] Figure 9 The preliminary optimized transmittance image in an embodiment;

[0055] Figure 10 The final transmittance image in an embodiment;

[0056] Figure 11 The schematic diagram of the comparison between the original foggy scene image and the fog-free image in an embodiment: Figure 11 (a) is the original foggy scene image, Figure 11 (b) is the fog-free image. Detailed implementation manners

[0057] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0058] First, analyze the root causes of image degradation in foggy environments, which mainly come from two points: one is the image blur caused by stray light formed by the scattering of atmospheric light during transmission; the other is the absorption and scattering of the reflected light of the target by the gel particles suspended in the air, resulting in a decrease in image brightness and contrast. According to the polarization imaging mechanism, when light scatters during propagation, its polarization state changes accordingly, and the atmospheric light changes from unpolarized light to partially polarized light. The polarization state of the target light depends not only on the scattering effect with the suspended particles in the atmosphere during its propagation but also on the material of the target itself, surface roughness, and other information. Therefore, the scene polarization information received by the detector contains not only the polarization information of the atmospheric light but also the polarization information of the target. Therefore, the polarization information of the target cannot be ignored in the process of restoring the fog-free image.

[0059] In one embodiment, as Figure 1 shown, a method for image defogging based on polarization characteristics and atmospheric transmission model is provided, including the following steps:

[0060] Step S1, acquire initial polarization images at four angles according to a four-channel optical polarization detector, calculate the Stokes vector based on the initial polarization images, and calculate the scene polarization degree image, scene polarization angle image, and scene light intensity image based on the Stokes vector.

[0061] It can be understood that the Stokes vector can be represented as a 4-row and 1-column matrix, and the four parameters in it are the intensity value dimensions of different parameters. The initial polarization characteristics of the initial polarization images can be intuitively represented by the Stokes vector, and the scene polarization degree image, scene polarization angle image, and scene light intensity image can be calculated based on the polarization characteristics in the form of the Stokes vector.

[0062] Step S2, perform joint segmentation on the region to be segmented according to the gradient information and light intensity information in the scene polarization angle image and scene light intensity image to obtain the sky region. Calculate the atmospheric polarization degree and the atmospheric light intensity at infinity based on the mask of the sky region, the scene polarization degree image, and the scene light intensity image; calculate the polarization difference image based on the scene polarization degree image and the scene light intensity image.

[0063] It can be understood that since the regional gradient information reflects the degree of change of the image in that region, the gray value at the junction of the target and the sky will change significantly, and its gradient value will also change accordingly. In the sky region and the target region, the gray value changes relatively smoothly, and the corresponding gradient change is also small. Therefore, in theory, the sky region can be segmented according to the gradient information. Performing segmentation by combining light intensity information on the basis of gradient information can prevent interference caused by noise changes such as high reflection in the target region and further improve the accuracy of the sky region segmentation result.

[0064] It can be understood that since both the scene polarization degree image and the scene light intensity image can be represented by dual-angle images, the polarization difference image can be directly obtained from the scene polarization degree image and the scene light intensity image obtained by solving the Stokes vector, avoiding the cumbersome process of calculating the dual-angle image, simplifying the algorithm complexity, and reducing the running time of the algorithm.

[0065] Step S3: Use the Gaussian blur algorithm to filter the initial polarization image to obtain the atmospheric polarization image, calculate the target polarization image based on the atmospheric polarization image, and calculate the target polarization degree image based on the target polarization image.

[0066] It can be understood that the Gaussian blur algorithm is a filtering method that distributes weights according to the normal distribution. Due to the continuity of the pixel points in the image, the closer the images are to each other, the higher their correlation, and the farther apart, the lower the correlation. Filtering the initial polarization image according to the Gaussian blur algorithm can ensure image smoothing filtering while maintaining the edge features of the image, thereby improving the accuracy of the estimated atmospheric polarization image.

[0067] It can be understood that the specific method of calculating the target polarization degree image based on the target polarization image is the same as that of calculating the scene polarization degree image based on the initial polarization image, which is to calculate the Stokes vector from the polarization images at four angles and then calculate the corresponding polarization degree image based on the Stokes vector.

[0068] Step S4: Calculate based on the scene light intensity image, atmospheric polarization degree, atmospheric light intensity at infinity, polarization difference image, and target polarization degree image to obtain the initial transmittance image. Determine the sky region of the initial transmittance image through the mask of the sky region, optimize and correct the initial transmittance image based on the sky region of the initial transmittance image and the guided filter algorithm to obtain the final transmittance image, and reconstruct the fog-free image based on the final transmittance image.

[0069] It can be understood that there are still many noise points in the sky region of the initial transmittance image, which need to be further optimized. Determine the sky region of the initial transmittance image through the mask of the sky region, and optimize the transmittance of the sky region of the initial transmittance image in combination with the sky region of the initial transmittance image to solve the problem of halos in the restored image caused by too strong scene light intensity; use the guided filter to deeply optimize the transmittance, smooth and reduce the noise points caused by the fragmentation of the transmittance image due to sky segmentation problems and inaccurate estimation of the target polarization degree image, avoiding the interference of noise in the atmosphere and the high-reflection part of the target area, obtaining the final transmittance image, and reconstructing the fog-free image based on the final transmittance image

[0070] In one embodiment, the Stokes vector is calculated based on the initial polarization image, and the scene polarization degree image, the scene polarization angle image, and the scene light intensity image are calculated based on the Stokes vector, including:

[0071] The Stokes vector is calculated based on the initial polarization image, expressed as

[0072] S = [S0, S1, S2, S3] T

[0073]

[0074] where S0 represents the total light intensity received by the four-channel optical polarization detector, S1 represents the intensity difference between the horizontal component light intensity and the vertical component light intensity, S2 represents the intensity difference between the 45° linear polarization and the 135° linear polarization, S3 represents the intensity difference between the right-handed circular polarization and the left-handed circular polarization. Usually, the component of circular polarization light in the polarization detection imaging system is small, and S3 is often ignored. I(0), I(45), I(90), and I(135) represent the initial polarization images at four angles of 0°, 45°, 90°, and 135° collected by the four-channel optical polarization detector. Among them, the initial polarization images at the four angles are as Figure 2 shown Figure 2 (a) is the initial polarization image at 0°, Figure 2 (b) is the initial polarization image at 45°, Figure 2 (c) is the initial polarization image at 90°, Figure 2 (d) is the initial polarization image at 135°;

[0075] Based on the polarization characteristics in the form of the Stokes vector, the scene polarization degree image P, the scene polarization angle image AOP, and the scene light intensity image I are further calculated and respectively expressed as

[0076]

[0077]

[0078] I = S0

[0079] where the scene polarization degree (DOP) image and the scene polarization angle (AOP) image are as Figure 3 (a) and Figure 3 (b) shown

[0080] In one embodiment, the region to be segmented is jointly segmented according to the gradient information and the light intensity information in the scene polarization angle image and the scene light intensity image to obtain the sky region, including:

[0081] Extract the gradient information of the region to be segmented in the scene polarization angle image and the scene intensity image, and set the first threshold and the second threshold to judge the change of the gradient information;

[0082] Extract the intensity information of the region to be segmented in the scene intensity image, and set the third threshold to judge the change of the intensity information,

[0083] Perform joint segmentation on the region to be segmented according to the first threshold, the second threshold and the third threshold to obtain the sky region.

[0084] Specifically, the Sobel operator is used to extract the gradient values of the region to be segmented in the scene polarization angle image and the scene intensity image. The Sobel operator uses two 3*3 matrices as the detection templates in the horizontal and vertical directions respectively. After performing convolution operations on the image with the templates, the horizontal direction gradient amplitude and the vertical direction gradient amplitude can be obtained. Then, according to the 1 / 2 power of the sum of the squares of the horizontal direction gradient amplitude and the vertical direction gradient amplitude, the gradient value of the region to be segmented is obtained, which is expressed as

[0085]

[0086]

[0087]

[0088] where G x represents the horizontal direction gradient amplitude, G y represents the vertical direction gradient amplitude, and A0 represents the original image to be segmented.

[0089] In one embodiment, when the gradient information of the region to be segmented in the scene polarization angle image is lower than the first threshold, the gradient information of the region to be segmented in the scene intensity image is lower than the second threshold, and the intensity information of the region to be segmented in the scene intensity image is greater than the third threshold, it is determined that the region to be segmented is the sky region; otherwise, it is determined that the region to be segmented is the target region;

[0090] According to the binarization operation, the pixel points in the sky region are assigned a value of 1, and the pixel points in the target region are assigned a value of 0. The sky region is represented as

[0091] Sky(x,y) = 1((G AOP (x,y) < th1) and (G I (x,y) < th2) and (I(x,y) > th3)

[0092] where G AOP (x,y) represents the gradient information of the region to be segmented in the scene polarization angle image, G I(x, y) represents the gradient information of the region to be segmented in the scene light intensity image, I(x, y) represents the light intensity information of the region to be segmented in the scene light intensity image, th1, th2, and th3 respectively represent the first threshold, the second threshold, and the third threshold, AOP represents the scene polarization angle image, and I represents the scene light intensity image.

[0093] It can be understood that by jointly segmenting the region to be segmented with the first threshold, the second threshold, and the third threshold, the sky region can be obtained, and by adjusting the three thresholds, the noise in the sky region and the situation where the target region is not fully included during the segmentation process can be eliminated. When the first threshold and the second threshold are fixed, adjusting the third threshold can expand the size of the target region, but at the same time, some sky regions with intensity values close to the target will be added. However, since the segmented sky region is mainly for calculating the degree of atmospheric polarization, the atmospheric light intensity at infinity, and the optimization of the later transmittance, and the added sky region is near the target rather than the infinity region, it does not affect the subsequent calculation results.

[0094] In one embodiment, calculations are performed based on the mask of the sky region, the scene polarization degree image, and the scene light intensity image to obtain the degree of atmospheric polarization and the atmospheric light intensity at infinity; calculations are performed based on the scene polarization degree image and the scene light intensity image to obtain the polarization difference image, including:

[0095] Multiply the mask of the sky region by the scene polarization degree image to obtain the scene polarization degree image of the sky region. By statistically analyzing the polarization degrees of all pixel points in the scene polarization degree image of the sky region, select the polarization degree with the most occurrences as the degree of atmospheric polarization P a , and the statistical result of the degree of atmospheric polarization is as Figure 4 shown;

[0096] Multiply the mask of the sky region by the scene light intensity image to obtain the scene light intensity image of the sky region. Sort the intensity values of all pixel points in the scene light intensity image of the sky region in descending order, and select the average value of the intensity values of the top one hundred pixel points as the atmospheric light intensity A at infinity inf ;

[0097] Calculations are performed based on the scene polarization degree image and the scene light intensity image to obtain the polarization difference image I d , which is expressed as

[0098] I d =P*I = P*(I(0)+I(90))

[0099] Among them, due to the characteristic of smooth distribution of the atmosphere, the degree of atmospheric polarization and the atmospheric light intensity at infinity representing the characteristics of atmospheric light can be considered as global parameters.

[0100] It can be understood that statistical methods are used in the calculation process of atmospheric polarization degree and atmospheric light intensity at infinity, which effectively avoids the estimation errors of atmospheric polarization degree and atmospheric light intensity at infinity caused by interference in the sky area.

[0101] It can be understood that calculating the polarization difference image by using the scene polarization degree obtained by Stokes vector calculation and the initial polarization image obtained by direct acquisition avoids the tedious process of calculating dual-angle images, simplifies the algorithm complexity, and reduces the algorithm running time.

[0102] In one embodiment, the initial polarization image is filtered using a Gaussian blur algorithm to obtain an atmospheric polarization image, the target polarization image is calculated based on the atmospheric polarization image, and the target polarization degree image is calculated based on the target polarization image.

[0103] The Gaussian blur algorithm is used to filter the initial polarization images I(0), I(45), I(90) and I(135) at four angles to obtain the atmospheric polarization images A(0), A(45), A(90) and A(135) at four angles. The filtering formula is expressed as

[0104] A=max(min(aB,I′),0)

[0105] B=Gauss(I′)-Gauss(|I′-Gauss(I′)|)

[0106] Where I′ represents the image to be filtered, B represents the intermediate value of the initial filtering, A represents the final atmospheric light intensity, and a represents the adjustment parameter used to control the image recovery intensity. The atmospheric polarization images at four angles, A(0), A(45), A(90), and A(135), are respectively as follows: Figure 5 (a) Figure 5 (b) Figure 5 (c) and Figure 5 (d)

[0107] The target polarization images D(0), D(45), D(90) and D(135) at four angles are calculated based on the atmospheric polarization images A(0), A(45), A(90) and A(135) at four angles, as shown in the following table: Figure 6 (a) Figure 6 (b) Figure 6 (c) and Figure 6 (d) shows that the target light intensity of the target polarization image at four angles is expressed as

[0108] D = IA;

[0109] The Stokes vector is calculated based on the target light intensity of the target polarization images at four angles to represent the target polarization characteristics, and the target polarization degree image P is calculated based on the target polarization characteristics in the form of the Stokes vector. d , such as Figure 7 shown.

[0110] In one embodiment, an initial transmittance image is obtained by calculating based on the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image, and the target polarization degree image, including:

[0111] An initial transmittance image is obtained by calculating based on the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image, and the target polarization degree image, which is expressed as:

[0112]

[0113] where P d represents the target polarization degree image, I d represents the polarization difference image, P a represents the atmospheric polarization degree, A inf represents the atmospheric light intensity at infinity. As shown in 8, there are still many noise points in the sky area of the initial transmittance image, and further optimization is needed. Figure 8 It can be seen from

[0114] In one embodiment, the sky area of the initial transmittance image is determined by a mask of the sky area, and the initial transmittance image is optimized and corrected according to the sky area of the initial transmittance image and the guided filter algorithm to obtain the final transmittance image. The fog-free image is reconstructed according to the final transmittance image, including:

[0115] The sky area of the initial transmittance image is determined by a mask of the sky area. When the proportion of the sky area of the initial transmittance image exceeds half of the initial transmittance image, the initial transmittance image is optimized according to the sky area of the initial transmittance image to obtain a preliminary optimized transmittance image as shown in Figure 9 , and the preliminary optimized transmittance image is corrected according to the guided filter algorithm to obtain a final transmittance image as shown in Figure 10 ; when the proportion of the sky area of the initial transmittance image does not exceed half of the initial transmittance image, the initial transmittance image is optimized according to the guided filter algorithm to obtain the final transmittance image;

[0116] Specifically, when the proportion of the sky area of the initial transmittance image exceeds half of the initial transmittance image, random assignment processing is performed on the sky area of the initial transmittance image to obtain a preliminary optimized transmittance image with the transmittance maintained within the range of 0.05 ± 0.01;

[0117] According to the atmospheric transmission model, combined with the final transmittance image t final , the atmospheric light intensity A at infinity inf and the light intensity image of scene I are calculated, and the haze-free image is reconstructed and expressed as

[0118]

[0119] Comparative analysis shows that Figure 11 the original hazy scene image shown in Figure 11 (a) and the haze-free image shown in

[0120] It can be found that according to the algorithm proposed in this application, the imaging quality of the optical imaging system in foggy weather can be effectively improved, and the restoration of foggy images can be realized.

[0121] It should be understood that although Figure 1 the steps in the flowchart of Figure 1 are shown in sequence according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover,

[0122] In one embodiment, an image dehazing device based on polarization characteristics and atmospheric transmission model is provided, including: an initial polarization image preliminary processing module, a sky region joint segmentation module, and a target polarization degree image acquisition module, where:

[0123] The initial polarization image preliminary processing module is used to obtain initial polarization images at four angles according to the acquisition of a four-channel optical polarization detector, calculate the Stokes vector based on the initial polarization images, and calculate the scene polarization degree image, the scene polarization angle image, and the scene light intensity image based on the Stokes vector;

[0124] The sky region joint segmentation module is used to perform joint segmentation on the region to be segmented according to the gradient information and light intensity information in the scene polarization angle image and the scene light intensity image to obtain the sky region, calculate the atmospheric polarization degree and the atmospheric light intensity at infinity according to the mask of the sky region, the scene polarization degree image, and the scene light intensity image; calculate the polarization difference image according to the scene polarization degree image and the scene light intensity image;

[0125] The target polarization degree image acquisition module is used to filter the initial polarization image by using the Gaussian blur algorithm to obtain the atmospheric polarization image, calculate the target polarization image based on the atmospheric polarization image, and calculate the target polarization degree image based on the target polarization image;

[0126] The haze-free image reconstruction module is used to calculate the initial transmittance image according to the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image, and the target polarization degree image, determine the sky region of the initial transmittance image through the mask of the sky region, optimize and correct the initial transmittance image according to the sky region of the initial transmittance image and the guided filter algorithm to obtain the final transmittance image, and reconstruct the haze-free image according to the final transmittance image.

[0127] For the specific limitations of the image dehazing device based on polarization characteristics and the atmospheric transmission model, reference can be made to the limitations of the image dehazing method based on polarization characteristics and the atmospheric transmission model in the above text, which will not be elaborated here. Each module in the above-mentioned image dehazing device based on polarization characteristics and the atmospheric transmission model can be implemented in whole or in part by software, hardware, and their combination. The above-mentioned modules can be embedded in the processor in the computer device in the form of hardware or be independent of it, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0128] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0129] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the appended claims.

Claims

1. An image defogging method based on polarization characteristics and an atmospheric transmission model, characterized in that, The method includes: Acquire initial polarization images at four angles according to a four-channel optical polarization detector, calculate the Stokes vector based on the initial polarization images, and calculate a scene polarization degree image, a scene polarization angle image, and a scene light intensity image based on the Stokes vector; Perform joint segmentation on the region to be segmented according to the gradient information and light intensity information in the scene polarization angle image and the scene light intensity image to obtain a sky region, calculate the atmospheric polarization degree and the atmospheric light intensity at infinity according to the mask of the sky region, the scene polarization degree image, and the scene light intensity image; calculate a polarization difference image according to the scene polarization degree image and the scene light intensity image; Perform filtering processing on the initial polarization images by using a Gaussian blur algorithm to obtain an atmospheric polarization image, calculate a target polarization image according to the atmospheric polarization image, and calculate a target polarization degree image according to the target polarization image; Calculate an initial transmittance image according to the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image, and the target polarization degree image, determine the sky region of the initial transmittance image through the mask of the sky region, optimize and correct the initial transmittance image according to the sky region of the initial transmittance image and a guided filtering algorithm to obtain a final transmittance image, and reconstruct a fog-free image according to the final transmittance image.

2. The method according to claim 1, wherein Calculating the Stokes vector according to the initial polarization images includes: Calculating the Stokes vector according to the initial polarization images, expressed as S = [S0, S1, S2, S3] T where S0 represents the total light intensity received by the four-channel optical polarization detector, S1 represents the intensity difference between the horizontal component light intensity and the vertical component light intensity, S2 represents the intensity difference between the 45° linear polarization and the 135° linear polarization, S3 represents the intensity difference between the right-handed circular polarization and the left-handed circular polarization, and I(0), I(45), I(90), and I(135) represent the initial polarization images at four angles.

3. The method according to claim 1, characterized in that, Performing joint segmentation on the region to be segmented according to the gradient information and light intensity information in the scene polarization angle image and the scene light intensity image to obtain a sky region includes: Extract the gradient information of the region to be segmented in the scene polarization angle image and the scene light intensity image, and set a first threshold and a second threshold to judge the change of the gradient information; Extract the light intensity information of the region to be segmented in the scene light intensity image, and set a third threshold to judge the change of the light intensity information, Perform joint segmentation on the region to be segmented according to the first threshold, the second threshold, and the third threshold to obtain a sky region.

4. The method according to claim 3, characterized in that, Performing joint segmentation on the region to be segmented according to the first threshold, the second threshold, and the third threshold to obtain a sky region includes: When the gradient information of the region to be segmented in the scene polarization angle image is lower than the first threshold, the gradient information of the region to be segmented in the scene light intensity image is lower than the second threshold, and the light intensity information of the region to be segmented in the scene light intensity image is greater than the third threshold, judge that the region to be segmented is a sky region; otherwise, judge that the region to be segmented is a target region; Assign the pixel points in the sky region to 1 and the pixel points in the target region to 0 according to the binarization operation, where the sky region is represented as Sky(x,y) = 1((G AOP (x,y) < th1) and (G I (x,y) < th2) and (I(x,y) > th3) Among them, G AOP (x, y) represents the gradient information of the region to be segmented in the scene polarization angle image, G I (x, y) represents the gradient information of the region to be segmented in the scene light intensity image, I(x, y) represents the light intensity information of the region to be segmented in the scene light intensity image, th1, th2, and th3 respectively represent the first threshold, the second threshold, and the third threshold, AOP represents the scene polarization angle image, and I represents the scene light intensity image.

5. The method according to any one of claims 1 to 4, characterized in that, Perform calculations based on the mask of the sky region, the scene polarization degree image, and the scene light intensity image to obtain the atmospheric polarization degree and the atmospheric light intensity at infinity, including: Multiply the mask of the sky region by the scene polarization degree image to obtain the sky region scene polarization degree image. By statistically analyzing the polarization degrees of all pixel points in the sky region scene polarization degree image, select the polarization degree that appears most frequently as the atmospheric polarization degree; Multiply the mask of the sky region by the scene light intensity image to obtain the sky region scene light intensity image. Sort the intensity values of all pixel points in the sky region scene light intensity image in descending order, and select the average value of the intensity values of the top 100 pixel points as the atmospheric light intensity at infinity.

6. The method according to claim 5, wherein Perform filtering processing on the initial polarization image using the Gaussian blur algorithm to obtain the atmospheric polarization image, including: Perform filtering processing on the initial polarization images at four angles using the Gaussian blur algorithm to obtain the atmospheric polarization images at four angles. The filtering formula is expressed as A = max(min(aB, I′), 0) B = Gauss(I′) - Gauss(|I′ - Gauss(I′)|) where I′ represents the image to be filtered, B represents the intermediate filtering value after initial filtering, A represents the finally obtained atmospheric light intensity, and a represents the adjustment parameter.

7. The method according to claim 6, characterized in that Perform calculations based on the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image, and the target polarization degree image to obtain the initial transmittance image, which is expressed as: Among them, P d represents the target degree of polarization image, I d represents the polarization difference image, P a represents the atmospheric degree of polarization, A inf represents the atmospheric light intensity at infinity.

8. The method according to claim 1, wherein Determine the sky region of the initial transmittance image through the mask of the sky region, and optimize and correct the initial transmittance image according to the sky region of the initial transmittance image and the guided filtering algorithm to obtain the final transmittance image, including: Determine the sky region of the initial transmittance image through the mask of the sky region. When the proportion of the sky region of the initial transmittance image exceeds half of the initial transmittance image, optimize the initial transmittance image according to the sky region of the initial transmittance image to obtain a preliminary optimized transmittance image, and correct the preliminary optimized transmittance image according to the guided filtering algorithm to obtain the final transmittance image; when the proportion of the sky region of the initial transmittance image does not exceed half of the initial transmittance image, optimize the initial transmittance image according to the guided filtering algorithm to obtain the final transmittance image.

9. The method according to claim 8, characterized in that, When the proportion of the sky region of the initial transmittance image exceeds half of the initial transmittance image, optimize the initial transmittance image according to the sky region of the initial transmittance image to obtain a preliminary optimized transmittance image, including: When the proportion of the sky region of the initial transmittance image exceeds half of the initial transmittance image, perform random assignment processing on the sky region of the initial transmittance image to obtain a preliminary optimized transmittance image with the transmittance maintained within the range of 0.05 ± 0.

01.

10. An image defogging device based on polarization characteristics and an atmospheric transmission model, characterized in that, The device includes: Initial polarization image preliminary processing module, which is used to acquire initial polarization images at four angles according to the four-channel optical polarization detector, calculate the Stokes vector based on the initial polarization images, and calculate the scene polarization degree image, the scene polarization angle image and the scene light intensity image based on the Stokes vector; Sky region joint segmentation module, which is used to perform joint segmentation on the region to be segmented according to the gradient information and light intensity information in the scene polarization angle image and the scene light intensity image to obtain the sky region, calculate the atmospheric polarization degree and the atmospheric light intensity at infinity according to the mask of the sky region, the scene polarization degree image and the scene light intensity image; calculate the polarization difference image according to the scene polarization degree image and the scene light intensity image; Target polarization degree image acquisition module, which is used to filter the initial polarization image by using the Gaussian blur algorithm to obtain the atmospheric polarization image, calculate the target polarization image according to the atmospheric polarization image, and calculate the target polarization degree image according to the target polarization image; Fog-free image reconstruction module, which is used to calculate the initial transmittance image according to the scene light intensity image, the atmospheric polarization degree, the atmospheric light intensity at infinity, the polarization difference image and the target polarization degree image, determine the sky region of the initial transmittance image through the mask of the sky region, optimize and correct the initial transmittance image according to the sky region of the initial transmittance image and the guided filter algorithm to obtain the final transmittance image, and reconstruct the fog-free image according to the final transmittance image.

Images