Smoke removing method for atmospheric scattering model image of smelting workshop

Abstract

The invention provides a method for removing smoke from an atmospheric scattering model image in a smelting workshop, aiming at the problems of image degradation and the like caused by a large amount of smoke, water mist and other suspended particles in the smelting workshop. In order to better estimate a real atmospheric light value, an initial atmospheric light value is solved through a simple linear iterative clustering segmentation method, the initial atmospheric light value is refined by adopting fast guided filtering, and meanwhile, atmospheric light and an original smoke image are corrected by utilizing a self-adaptive gamma function, so that the real atmospheric light value is estimated. Respectively obtaining final atmospheric light and an optimized smoke image; the transmissivity is estimated through the optimized color attenuation prior model, and finally the smoke-free image is recovered according to the atmospheric scattering model. Experimental results show that the method can reduce the smoke concentration in the image, reduce the detail loss of the image, averagely reduce the square-average error by 66.2%, averagely improve the peak signal-to-noise ratio by 30.5%, and averagely improve the structural similarity by 48.6%.

Description

Technical Field

[0001] This invention addresses the image degradation caused by a large amount of suspended particles such as smoke and water mist in smelting workshops by proposing a method for removing smoke and dust from atmospheric scattering model images of smelting workshops. Background Technology

[0002] With the rapid development of computer vision technology, significant progress has been made in industrial informatization and intelligentization. Video surveillance technology, as an important component of intelligent manufacturing, is increasingly widely used in industrial fields such as non-ferrous metal smelting. To improve production efficiency and ensure production safety, more and more factories are installing network cameras in their production workshops to collect real-time production information. However, during the smelting production process, due to the large amount of smoke, dust, water vapor, and other pollutants generated in the high-temperature working environment, monitoring video images often suffer from problems such as blurred details, reduced contrast, and weakened visibility. This not only affects the safety monitoring effect of the production process but also limits the application efficiency of video-based remote control of equipment and intelligent video recognition technology. Therefore, developing effective image processing methods to remove interference factors such as smoke and dust from smelting workshop monitoring images is of great significance for improving the reliability and intelligence level of monitoring systems.

[0003] Current image dehazing techniques are mainly divided into three categories: image enhancement methods, image restoration methods, and deep learning-based methods. Image enhancement techniques focus on improving image contrast and local features, without directly considering the fog formation mechanism and its impact on the image. Although this type of method can effectively reduce the fog effect while preserving image details, its versatility is limited. Specific parameters need to be adjusted for different images, and it may cause color distortion under high fog concentration conditions.

[0004] Image restoration methods are based on physical models, estimating the true values ​​of parameters by inversely analyzing the image degradation process, and thus restoring the original fog-free image. Deep learning-based methods rely on deep neural networks to learn the mapping relationship between foggy and fog-free images. Although these methods can effectively remove fog, they require a large number of paired foggy and fog-free image datasets for training. These datasets are often difficult to obtain and have long training cycles, thus limiting their industrial applications. Most existing technologies focus on image defogging in outdoor environments and fail to fully consider the influence of complex factors such as multiple light sources and uneven smoke and dust distribution in complex imaging environments, making them unsuitable for smoke and dust removal needs in specific industrial environments such as smelting workshops.

[0005] In recent years, existing defogging technologies have limitations in specific application scenarios, especially for smoke and dust treatment in industrial environments. In view of this, this invention proposes a smoke and dust removal method for atmospheric scattering model images in smelting workshop environments. It aims to improve the accuracy of atmospheric light value and transmittance estimation in complex environments and effectively solve the problems of halo effect and color distortion in bright areas. Experimental results have confirmed the significant effectiveness of the method of this invention in smoke and dust removal. Summary of the Invention

[0006] This invention provides an image smoke and dust removal method based on an atmospheric scattering model, the flowchart of which is shown below. Figure 1 As shown, this method aims to solve the problem of image quality degradation caused by factors such as smoke and water mist in smelting workshops. The method optimizes atmospheric light value estimation through a simple linear iterative clustering segmentation method and fast guided filtering, and uses an adaptive gamma function for image correction. Based on the optimized color attenuation prior model, the transmittance is calculated, and finally, a smoke-free image is recovered.

[0007] This invention is achieved through the following technical solution:

[0008] A method for removing smoke and dust from images based on an atmospheric scattering model is proposed. This method optimizes atmospheric light value estimation through a simple linear iterative clustering segmentation method and a fast guided filter, and uses an adaptive gamma function for image correction. Transmittance is calculated based on an optimized color attenuation prior model, ultimately restoring a smoke-free image. The original image of an aluminum electrolysis plant with smoke and dust is shown below. Figure 2 As shown;

[0009] The atmospheric scattering model is a model applied in image restoration and dehazing methods. It describes the degradation process of foggy images by analyzing the influence of various ambient lights during the object imaging process. In outdoor environments, atmospheric light is often regarded as a global constant value. Traditional methods select the pixels with the highest brightness intensity in the dark channel image and determine the pixel corresponding to the maximum brightness value as the atmospheric light value. However, this method is only suitable for scenes with uniform lighting. In indoor smelting workshops, due to the uneven light source, directly using this method may result in local overexposure or halo phenomena in the image after removing smoke and dust.

[0010] As an optimization, a new method for estimating atmospheric light values ​​is proposed by combining the superpixel segmentation method of simple linear iterative clustering.

[0011] As an optimization, the SLIC method is used to segment regions with similar atmospheric light values ​​in the image. The image after atmospheric light value processing is shown below. Figure 3 As shown, the atmospheric light value of each pixel in each region is the average value of all pixels in that region;

[0012] As an optimization, the selected pixel feature vectors, taking into account the characteristics of smoke and dust images, include saturation (S), brightness (V), and position coordinates (x,y).

[0013] As an optimization, the initial atmospheric light value estimation process is as follows:

[0014] A1. Select the feature vector of the pixel;

[0015] A2. Initialize cluster centers. Assuming the total number of image pixels is N, the number of clusters is set to K, and the adjacent distance is L, then we have:

[0016] A3. Move the initial cluster center to the position with the smallest ladder within the 3*3 neighborhood of the cluster center;

[0017] A4. Detect the similarity between cluster centers within a 2L*2L neighborhood of each pixel, and assign the cluster label with the highest similarity to the pixel to that pixel;

[0018] A5. Repeat the above process to iteratively update the cluster centers until the error converges;

[0019] A6. Calculate the average value of each cluster and assign the average value of each cluster to each pixel in each cluster.

[0020] The simple linear iterative clustering superpixel segmentation method can accurately estimate the true atmospheric light value of the approximate region; however, the image suffers from block distortion, which is due to the excessively coarse initial atmospheric light value, resulting in uneven transitions in the blocky regions.

[0021] As an optimization, the initial atmospheric light value was optimized, and the image before optimization (excluding smoke and dust) is shown below. Figure 4 As shown, in order to obtain a more natural smoke and dust removal effect, this invention uses fast guided filtering to refine the initial atmospheric light value;

[0022] The transmittance estimation can be obtained by calculating the transmittance after obtaining the atmospheric scattering coefficient and scene depth; the saturation, brightness and gradient features of the smoke and dust image in the smelting workshop approximately conform to the color attenuation model, and the scene depth in this area can be linearly represented by brightness, saturation and gradient.

[0023] As an optimization, the specific expression for the scene depth of the smoke and dust image is:

[0024] d = θ1V + θ2S + θ3G + ε;

[0025] Where θ1, θ2, and θ3 are linear coefficients; d is the image scene depth; V is the brightness; S is the saturation; G is the gradient; and ε is the error coefficient.

[0026] The aforementioned smoke and dust image correction: In an indoor smelting workshop, multiple artificial light sources and uneven smoke and dust distribution cause the image acquisition equipment to receive excessive scattered and direct light information, resulting in image degradation; in order to reduce the impact of excessive light and improve details in dark areas, it is necessary to appropriately weaken the brightness component in the overly bright areas of the original smoke and dust image and make appropriate compensation in the darker areas.

[0027] As an optimization, in order to reduce the impact of excessive illumination and improve details in dark areas, it is necessary to appropriately reduce the brightness component in the overly bright areas of the original smoke image and make appropriate compensation in the darker areas.

[0028] This invention also proposes a preprocessing method for correcting smoke and dust images using an adaptive gamma function. This method dynamically adjusts the brightness value based on the characteristics of each pixel, achieving precise control over image correction. The specific process is as follows:

[0029] B1. Convert the smoke image I from the RGB color model to the HSV color model in order to process the luminance component;

[0030] B2. Using the luminance component itself as a guide map, a lighting map is obtained through fast guided filtering.

[0031] B3. Perform gamma correction on the luminance component;

[0032] B4. Re-integrate the corrected brightness P with the original hue and saturation;

[0033] B5. Convert the fused image from the HSV color model to the RGB color model to obtain the corrected smoke and dust image. Attached Figure Description

[0034] The accompanying drawings, which are included to provide a further understanding of the embodiments of the present invention and constitute a part of this application, are not intended to limit the embodiments of the present invention; in the drawings:

[0035] Figure 1 This is a flowchart of the smoke and dust removal process of the present invention;

[0036] Figure 2 This is the original drawing of the electrolytic aluminum plant with smoke and dust, as described in this invention;

[0037] Figure 3 This is the original image in this invention after initial atmospheric light value processing;

[0038] Figure 4 This is the original image in this invention after optimization to remove smoke and dust;

[0039] Figure 5 This is the image after the original image in this invention has been refined by guided filtering;

[0040] Figure 6 This is the original image in this invention after gamma correction;

[0041] Figure 7 This is the image of the original image after optimization to remove smoke and dust in this invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0043] First, the atmospheric scattering model assumes that ambient light is a global constant, and represents the degraded foggy image as a linear superposition of the attenuation of light reflected from objects and the absorption and scattering of ambient light by suspended particles. The smoke removal method based on image restoration is based on this model. It compensates for the attenuation energy of reflected light and eliminates the part of ambient light involved in imaging, thereby restoring a clear image.

[0044] Based on the above imaging process of smoke and dust environment, the expression of the degraded smoke and dust image is as follows:

[0045] I = Jt + A(Et);

[0046] Where I is the received smoke and dust image; J is the restored image; t is the transmittance of the medium, which reflects the ability of atmospheric light to penetrate the medium and reach the acquisition device. The larger the value, the stronger the penetration ability; A is the global atmospheric light; and E is the identity matrix.

[0047] In the atmospheric scattering model, the medium transmittance is related to the atmospheric scattering coefficient and the scene depth, while the atmospheric scattering coefficient is related to the image brightness. The relationship is as follows:

[0048] t = e -βd ;

[0049] β=ae cV ;

[0050] Where β is the atmospheric scattering coefficient; d is the scene depth; ∨ is the brightness value of each pixel in the smoke image; a and c are constants, generally a = 1.2 and c = 0.25;

[0051] To address this, a novel method for estimating atmospheric light values ​​is proposed, which combines the superpixel segmentation method of Simple Linear Iterative Clustering (SLIC).

[0052] Specifically, the SLIC method is used to segment regions in the image with similar atmospheric light values, and the atmospheric light value of each pixel in each region is the average value of all pixels in that region.

[0053] The initial atmospheric light value estimation process is as follows:

[0054] A1. Select the feature vector of the pixel;

[0055] A2. Initialize cluster centers. Assuming the total number of image pixels is N, the number of clusters is set to K, and the adjacent distance is L, then we have:

[0056] A3. Move the initial cluster center to the position with the smallest ladder within the 3*3 neighborhood of the cluster center;

[0057] A4. Detect the similarity between cluster centers within a 2L*2L neighborhood of each pixel, and assign the cluster label with the highest similarity to the pixel to that pixel;

[0058] A5. Repeat the above process to iteratively update the cluster centers until the error converges;

[0059] A6. Calculate the average value of each cluster and assign the average value of each cluster to each pixel in each cluster.

[0060] The formula for detecting similarity is:

[0061]

[0062] In the formula d c d represents the color distance between the cluster center and its neighboring pixels. s S is the Euclidean distance between the cluster center and the neighboring pixels; D is the similarity between the cluster center and the pixels; S i V i S represents the saturation and brightness of pixel i, respectively; j V j x represents the saturation and brightness of the pixel at cluster center j within a 2L×2L neighborhood; j y j x represents the x and y coordinates of pixel i; i y i λ represents the x and y coordinates of the neighborhood cluster center i of the pixel; m is the compaction factor.

[0063] The SLIC superpixel segmentation method can estimate the true atmospheric light value of the approximate region relatively well; however, the image suffers from blocky distortion, which is due to the overly coarse initial atmospheric light value estimation, resulting in an uneven transition of blocky regions.

[0064] Therefore, the initial atmospheric light value is optimized to obtain a more natural smoke and dust removal effect; the present invention uses fast guided filtering to refine the initial atmospheric light value.

[0065] The transmittance estimation method can calculate the transmittance after obtaining the atmospheric scattering coefficient and scene depth. The saturation, brightness, and gradient characteristics of the smoke and dust image in the smelting workshop approximately conform to the color attenuation model, and the scene depth in this area can be linearly represented by brightness, saturation, and gradient.

[0066] d = θ1V + θ2S + θ3G + ε;

[0067] Where θ1, θ2, and θ3 are linear coefficients; d is the image scene depth; V is the brightness; S is the saturation; G is the gradient; and ε is the error coefficient.

[0068] The expression for the average scene depth of the estimated smoke image is:

[0069]

[0070] Where Ω i For (x) i y i () is the central window; and These represent the average brightness, average saturation, and average gradient within Ω, respectively; M is the number of windows.

[0071] In the indoor smelting workshop, multiple artificial light sources and uneven dust distribution cause the image acquisition equipment to receive excessive scattered and direct light information, resulting in a decrease in image quality. In order to reduce the impact of excessive light and improve the details in dark areas, it is necessary to appropriately weaken the brightness component in the overly bright areas of the original dust image and make appropriate compensation in the darker areas.

[0072] To address this, this invention proposes a preprocessing method for correcting smoke and dust images using an adaptive gamma function. This method dynamically adjusts the brightness value based on the characteristics of each pixel, achieving precise control over image correction. The specific process is as follows:

[0073] B1. Convert the smoke image I from the RGB color model to the HSV color model in order to process the luminance component;

[0074] B2. Using the luminance component itself as a guide map, a lighting map is obtained through fast guided filtering.

[0075] B3. Perform gamma correction on the luminance component;

[0076] B4. Re-integrate the corrected brightness P with the original hue and saturation;

[0077] B5. Convert the fused image from the HSV color model to the RGB color model to obtain the corrected smoke and dust image.

[0078] The expression for the gamma function is:

[0079] P = V γ ;

[0080]

[0081] Where P is the corrected brightness of the smoke image; V is the brightness component of the input image; γ is the gamma parameter used for brightness correction; l is the illumination component extracted from the input image; and ave is the average brightness of the illumination component.

[0082] The image smoke and dust removal method based on the atmospheric scattering model described above achieves effective removal of smoke and dust images by using a simple linear iterative clustering segmentation method and a fast guided filter to optimize atmospheric light value estimation, and then using an adaptive gamma function for image correction. This method includes steps such as initial atmospheric light value estimation, fast guided filter refinement, transmittance estimation, and adaptive gamma function correction. It can reduce the influence of scattered and direct light in images under complex indoor and outdoor lighting environments, improve image details and contrast, and restore smoke-free images. It is particularly suitable for environments with uneven light sources, such as smelting workshops, and effectively avoids overexposure and halo phenomena caused by traditional methods.

[0083] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for removing smoke and dust from images based on an atmospheric scattering model, characterized in that, Atmospheric light value estimation is optimized using a simple linear iterative clustering segmentation method and a fast guided filter, and image correction is performed using an adaptive gamma function. The atmospheric scattering model is a model applied in image restoration and dehazing methods. It describes the degradation process of foggy images by analyzing the influence of various ambient lights during the object imaging process.

2. According to the atmospheric scattering model described in claim 1, in outdoor environments, atmospheric light is often regarded as a global constant value. Traditional methods select the pixels with the highest brightness intensity in the dark channel image and determine the pixel corresponding to the maximum brightness value as the atmospheric light value. However, this method is only applicable to scenes with uniform lighting. In indoor smelting workshops, due to the uneven light source, directly using this method may result in local overexposure or halo phenomena in the image after removing smoke and dust. The initial atmospheric light value estimation process is as follows: A1. Select the feature vector of the pixel; A2. Initialize cluster centers. Assuming the total number of image pixels is N, the number of clusters is K, and the adjacent distance is L, then we have: A3. Move the initial cluster center to the position with the smallest ladder within the 3*3 neighborhood of the cluster center; A4. Detect the similarity between cluster centers within a 2L*2L neighborhood of each pixel, and assign the cluster label with the highest similarity to the pixel to that pixel; A5. Repeat the above process to iteratively update the cluster centers until the error converges; A6. Calculate the average value of each cluster and assign the average value of each cluster to each pixel in each cluster.

3. The simple linear iterative clustering superpixel segmentation method according to claim 1 can accurately estimate the true atmospheric light value of the approximate region. However, the image suffers from block distortion because the initial atmospheric light value is too coarse, resulting in uneven transitions in the blocky regions. To optimize the initial atmospheric light value, a more natural smoke and dust removal effect is obtained. Therefore, this invention employs fast guided filtering to refine the initial atmospheric light value.

4. The simplified smoke and dust image correction according to claim 1: In an indoor smelting workshop, multiple artificial light sources and uneven smoke and dust distribution cause the image acquisition equipment to receive excessive scattered and direct light information, resulting in image degradation. In order to reduce the influence of excessive light and improve the details in dark areas, it is necessary to appropriately weaken the brightness component in the overly bright areas of the original smoke and dust image and make appropriate compensation in the darker areas.

5. The preprocessing method for correcting smoke and dust images using the adaptive gamma function according to claim 1 dynamically adjusts the brightness value based on the characteristics of each pixel to achieve precise control over image correction. The specific process is as follows: B1. Convert the smoke image I from the RGB color model to the HSV color model in order to process the luminance component; B2. Using the luminance component itself as a guide map, a lighting map is obtained through fast guided filtering. B3. Perform gamma correction on the luminance component; B4. Re-integrate the corrected brightness P with the original hue and saturation; B5. Convert the fused image from the HSV color model to the RGB color model to obtain the corrected smoke and dust image.

6. The image smoke and dust removal method based on the atmospheric scattering model according to claims 1 to 5 achieves effective removal of smoke and dust images by using a simple linear iterative clustering segmentation method and a fast guided filter to optimize atmospheric light value estimation, and by using an adaptive gamma function for image correction. The method includes steps such as initial atmospheric light value estimation, fast guided filter refinement, transmittance estimation, and adaptive gamma function correction. It can reduce the influence of scattered and direct light in images under complex indoor and outdoor lighting environments, improve image details and contrast, and restore smoke and dust-free images. It is particularly suitable for environments with uneven light sources, such as smelting workshops, and effectively avoids overexposure and halo phenomena caused by traditional methods.

Images