Single image defogging method based on iterative transmissivity correction and related device

Abstract

The embodiment of the invention relates to the technical field of digital image processing and computer vision, and provides a single image defogging method based on iterative transmissivity correction and a related device, and the method comprises the steps: obtaining a to-be-processed foggy image; performing atmospheric light value estimation processing according to the to-be-processed foggy image to obtain a global atmospheric light value; on the basis of dark channel prior, according to the to-be-processed foggy image, transmissivity graph calculation processing is carried out, and an initial transmissivity graph is obtained; according to the to-be-processed foggy image and the global atmospheric light value, performing optimization iteration processing on the initial transmissivity graph to obtain a target transmissivity graph; based on the atmospheric scattering model, image defogging processing is carried out according to the to-be-processed foggy image, the global atmospheric light value and the target transmittance graph, so that a more accurate target restored image can be obtained, and the defogging processing accuracy of a single image can be effectively improved.

Description

Technical Field

[0001] This application relates to the fields of digital image processing and computer vision technology, specifically to a single-image dehazing method and related apparatus based on iterative transmittance correction. Background Technology

[0002] The performance of outdoor vision systems (such as autonomous driving, video surveillance, and drone aerial photography) is often affected by suspended particles in the atmosphere (fog, haze, smoke), leading to decreased contrast, color distortion, and loss of detail in captured images. Single-image dehazing aims to recover a clear scene from a single foggy input, which is a typical ill-conditioned inverse problem. Physically based methods, especially Dark Channel Prior (DCP), are widely used due to their solid physical foundation and good results. However, DCP fails in bright or low-texture areas such as the sky, water surfaces, and white objects because the "dark channel" value in these areas does not approach zero. This causes DCP to severely underestimate transmittance in these areas, resulting in obvious halo artifacts (unnatural bright / dark rings at object edges) and color distortion (overall darkening or local color cast). Therefore, how to effectively improve the accuracy of single-image dehazing has become an urgent problem to be solved. Summary of the Invention

[0003] This application provides a single-image dehazing method and related apparatus based on iterative transmittance correction, which can achieve more accurate automated marketing decisions.

[0004] A first aspect of this application provides a single-image dehazing method based on iterative transmittance correction, comprising: Acquire the foggy image to be processed; The atmospheric light value is estimated based on the foggy image to be processed to obtain the global atmospheric light value. Based on the dark channel prior, the transmittance map is calculated and processed according to the foggy image to be processed to obtain the initial transmittance map. Based on the foggy image to be processed and the global atmospheric light value, the initial transmittance map is optimized and iterated to obtain the target transmittance map; Based on the atmospheric scattering model, image dehazing is performed according to the foggy image to be processed, the global atmospheric light value, and the target transmittance map to obtain the target restored image.

[0005] A second aspect of this application provides a single-image dehazing device based on iterative transmittance correction, the single-image dehazing device based on iterative transmittance correction comprising: The acquisition unit is used to acquire the foggy image to be processed; The first processing unit is used to perform atmospheric light value estimation processing on the foggy image to be processed to obtain the global atmospheric light value. The second processing unit is used to perform transmittance map calculation based on the dark channel prior and the foggy image to be processed to obtain an initial transmittance map. The third processing unit is used to perform optimization and iterative processing on the initial transmittance map based on the foggy image to be processed and the global atmospheric light value to obtain the target transmittance map. The fourth processing unit is used to perform image dehazing based on the atmospheric scattering model, the foggy image to be processed, the global atmospheric light value, and the target transmittance map to obtain the target restored image.

[0006] A third aspect of this application provides a terminal including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute the step instructions as described in the first aspect of this application.

[0007] A fourth aspect of this application provides a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the first aspect of this application.

[0008] A fifth aspect of this application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the first aspect of this application. The computer program product may be a software installation package.

[0009] Implementing the embodiments of this application has the following beneficial effects: By acquiring a foggy image to be processed and performing atmospheric light value estimation based on the image, a global atmospheric light value is obtained. Further, based on a dark channel prior, a transmittance map calculation is performed on the foggy image to obtain an initial transmittance map. This initial transmittance map can then be optimized and iteratively processed based on the foggy image and the global atmospheric light value to obtain a target transmittance map. Subsequently, based on an atmospheric scattering model, image dehazing can be performed using the foggy image, the global atmospheric light value, and the target transmittance map, resulting in a more accurate target restoration image. This improves the accuracy of the dehazing process for a single image. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application 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 some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This application provides a flowchart illustrating a single-image dehazing method based on iterative transmittance correction as an embodiment of the present application. Figure 2A This application provides an overall flowchart for defogging in its embodiments; Figure 2B This application provides a schematic diagram of dark channel prior (DCP) failure for an embodiment of the present application (left 1: foggy image, left 2: transmittance map estimated by DCP, left 3: DCP defogging image, left 4: reference image). Figure 2C This application provides an example image comparing dehazing effects on the RESIDE dataset; Figure 2D This application provides an example image comparing dehazing effects on the NH-HAZE dataset for embodiments of the present application; Figure 3 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application; Figure 4 This application provides a schematic diagram of a single-image dehazing device based on iterative transmittance correction. Detailed Implementation

[0012] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0013] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0014] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0015] To better understand the single-image dehazing method based on iterative transmittance correction provided in this application, we will first briefly introduce methods based on physical models, such as the dark channel prior. Existing DCP (Dark Channel Correction) assumes that in most local blocks of hazy, non-sky areas, at least one color channel contains pixels with extremely low intensity. This prior can effectively estimate transmittance and atmospheric light. However, DCP fails in bright or low-texture areas such as the sky, water surfaces, and white objects, because the "dark channel" value in these areas does not approach zero. This leads to DCP severely underestimating transmittance in these areas, resulting in obvious halo artifacts (unnatural bright / dark rings at object edges) and color distortion (overall darkening or local color cast in the image), such as... Figure 2B As shown. To address the aforementioned issues with DCP, researchers have proposed several improvement schemes. For example, the proposed Color Attenuation Prior (CAP) quickly estimates transmittance by establishing a linear model of scene depth and pixel brightness and saturation, but it struggles to handle complex haze under non-uniform lighting. Alternatively, a proposed nonlocal dehazing method based on haze lines utilizes color clustering in the RGB space to improve transmittance consistency, but it is prone to producing halos at abrupt changes in depth of field. Improving DCP in the HSI color space (IDCP) alleviates the oversaturation problem but introduces color space dependency. Another approach is the proposed Multi-DCP framework, which fuses prior information from different scales and performs adaptive atmospheric light and gamma correction, improving color fidelity, but significantly increasing computational complexity. While these methods improve DCP performance to some extent, most still rely on manually designed heuristic rules, region segmentation, or multi-scale fusion, leaving room for improvement in terms of algorithm versatility, complexity, and efficiency. On the other hand, while deep learning-based dehazing methods (such as DehazeNet and AOD-Net) have achieved excellent visual quality, they heavily rely on large-scale paired training data, incur huge computational costs, and lack physical interpretability, making them difficult to deploy in embedded or real-time systems with strict requirements for latency, power consumption, and reliability. To address the aforementioned problems, this application provides a single-image dehazing method based on iterative transmittance correction. This method derives a physical correction term from an atmospheric scattering model, consisting of transmittance and the restored image, and designs an iterative update framework to automatically and progressively correct the underestimation of transmittance in bright or low-texture areas by traditional DCP methods. This effectively suppresses halo artifacts and improves color fidelity without requiring any region segmentation or deep learning training. Furthermore, this invention also provides an automatic hyperparameter optimization method based on robust statistical fusion of multiple indicators. This method scientifically and objectively determines key algorithm parameters, ensuring optimal performance and stability.

[0016] Please refer to Figure 2, which is a flowchart illustrating a single-image dehazing method based on iterative transmittance correction provided in this application embodiment. As shown in Figure 2, the single-image dehazing method based on iterative transmittance correction includes: S10: Obtain the foggy image to be processed.

[0017] The foggy image to be processed can refer to the original input image captured under adverse weather conditions such as fog, haze, or smoke, which exhibits reduced contrast, blurred details, and color distortion. For example, the foggy image to be processed can refer to images taken by outdoor surveillance cameras or autonomous driving vehicle cameras under adverse weather conditions. Optionally, the following description uses a foggy environment as an example and does not constitute a limitation on this application.

[0018] S20: Perform atmospheric light value estimation processing on the foggy image to be processed to obtain the global atmospheric light value.

[0019] Among them, atmospheric light value estimation processing can be understood as the process of analyzing the brightness distribution of a foggy image through algorithms (such as the high brightness pixel screening method based on the dark channel) to find the value of the global ambient light (i.e., atmospheric light value) formed by the scattering of suspended particles in the atmosphere.

[0020] Global atmospheric light value refers to the uniformly distributed background light intensity in a foggy environment. It should be noted that this global atmospheric light value is typically a value from the red, green, and blue (RGB) channels. This global atmospheric light value can be considered one of the core parameters of the atmospheric scattering model, determining the overall brightness baseline of the image.

[0021] Specifically, for the input color hazy image to be processed Global atmospheric light can be estimated using methods from the classic Dark Channel Prior (DCP). Calculate the dark channel of an image Select the pixels with the highest brightness in the dark channel (top 0.1%) from the original color hazy image to be processed. The average (or maximum) pixel value at the corresponding position is used as The estimate is as follows. Optional, the specific formula can be found below: in, It can refer to pixels The dark channel value at a given location can reflect the minimum color channel intensity value within a local area; It can refer to the color channel index, and the value can be red ( ),green( ) and blue ( ); Can refer to pixels A local image patch centered on Pixel position within; It can refer to the location Color channel The pixel intensity value; min() can be used to retrieve the minimum value.

[0022] S30: Based on the dark channel prior, the transmittance map is calculated and processed according to the foggy image to be processed to obtain the initial transmittance map.

[0023] Dark channel priors refer to the prior knowledge that, in most local image patches in fog-free, non-sky areas, at least one color channel (such as R / G / B) has pixels with extremely low intensity. This dark channel prior can be considered the core theoretical basis for calculating transmittance. Transmittance map calculation can be understood as the process of calculating the degree of light obstruction by fog at each pixel based on the dark channel prior and atmospheric light values, and presenting it in image form.

[0024] The initial transmittance map refers to the transmittance distribution map obtained through preliminary calculations using the dark channel prior. Each pixel value in this initial transmittance map represents the degree of fog transmission at that location; understandably, a pixel value closer to 1 indicates less fog, while a pixel value closer to 0 indicates denser fog.

[0025] Specifically, the initial transmittance map can be calculated based on the dark channel prior. For the specific formula, please refer to the following: in, It can refer to pixel-based prior estimation of the dark channel. The initial transmittance at the location can range from [0,1]. It should be noted that the smaller the initial transmittance value, the denser the fog. It can refer to a constant factor, which is usually set to 0.95. It can be used to retain a small amount of fog to maintain the sense of depth in the scene, thereby making the restoration result more natural. It can refer to the global atmospheric light value, or represent the atmospheric light intensity at infinity; It can refer to the pixels calculated in the aforementioned steps. The dark channel value at that location.

[0026] S40: Based on the foggy image to be processed and the global atmospheric light value, the initial transmittance map is optimized and iterated to obtain the target transmittance map.

[0027] The optimization and iterative processing can be understood as a process of gradually compensating for the estimation deviation of the initial transmittance map in areas such as the sky and highlights by updating the restored image, calculating the correction term, and correcting the transmittance map in a loop, thereby making the transmittance more accurate.

[0028] The target transmittance map refers to the unbiased transmittance distribution map obtained after the above iterative optimization and edge-preserving filtering. It can be understood that this target transmittance map can better match the actual fog distribution than the initial transmittance map, enabling more accurate dehazing in subsequent image dehazing processes.

[0029] S50: Based on the atmospheric scattering model, perform image dehazing processing according to the foggy image to be processed, the global atmospheric light value, and the target transmittance map to obtain the target restored image.

[0030] The atmospheric scattering model refers to a physical model used to describe the propagation of light in foggy weather. The core logic of this model is: a foggy image = light attenuated by fog in a fog-free image + the superposition of scattering from the atmosphere. This can be seen as the physical basis of dehazing algorithms. Image dehazing can be understood as a process based on the atmospheric scattering model, derived in reverse, using a foggy image, global atmospheric light values, and a target transmittance map to counteract the attenuation and scattering effects of fog, thereby restoring a clear image. The target restored image refers to the final clear, fog-free image output after dehazing. This can be understood as an image that has restored the contrast, details, and natural colors of the original scene.

[0031] Specifically, target transmittance maps can be used. Global atmospheric light value And the original foggy image (i.e., the foggy image to be processed). The final dehazed image, i.e., the target restored image, is calculated and output according to the following formula: , in, It can refer to the intensity value at pixel x in the final restored dehazed image (i.e., the target restored image); It can refer to the value at pixel x in the final calculated transmittance map (i.e., the target transmittance map); It can refer to the original foggy image, i.e., the foggy image to be processed; It can refer to the global atmospheric light value; It can be a small positive number (such as 0.05) to prevent the denominator from being zero, thus ensuring numerical stability; that is, to enhance numerical stability, the target transmittance in the denominator can be used. With a small positive number Take the maximum value.

[0032] This application calculates an initial transmittance map based on a priori calculation of the dark channel and performs iterative optimization processing on the initial transmittance map by combining the image to be processed (foggy) with global atmospheric light values. Based on the optimized target transmittance map, image dehazing is achieved, effectively solving the technical shortcomings of traditional dark channel priori dehazing methods, which suffer from insufficient accuracy in single transmittance estimation and are prone to halo artifacts and color distortion in complex areas such as the sky and highlights. Compared to existing technologies, this application does not rely on complex preprocessing procedures such as image segmentation and deep learning model training. It only relies on an atmospheric scattering physics model and iterative optimization logic to adaptively correct transmittance deviations. While accurately restoring image details and natural colors, it effectively controls computational complexity and possesses strong real-time processing capabilities. It can be directly deployed in resource-constrained scenarios such as autonomous driving vehicle terminals and security monitoring equipment. Furthermore, this application exhibits good adaptability and robustness to foggy images under different concentrations of fog and haze, achieving high-quality dehazing without manual parameter adjustment, significantly improving the practicality and applicability of single-image dehazing methods.

[0033] In this embodiment, by acquiring a foggy image to be processed and performing atmospheric light value estimation processing based on the foggy image to obtain global atmospheric light value, a transmittance map calculation can be further performed based on dark channel prior, based on the foggy image to be processed, to obtain an initial transmittance map. Then, based on the foggy image to be processed and the global atmospheric light value, the initial transmittance map can be optimized and iteratively processed to obtain a target transmittance map. Subsequently, based on an atmospheric scattering model, image dehazing processing can be performed based on the foggy image to be processed, the global atmospheric light value, and the target transmittance map, resulting in a more accurate target restoration image, which is beneficial for improving the accuracy of the dehazing process for a single image.

[0034] In one possible implementation, when optimizing the initial transmittance map, an initial haze-free image can first be reconstructed from the initial transmittance map and atmospheric light values. A correction term is then constructed based on this. Subsequently, the transmittance map is iteratively updated based on the correction term, the initial reconstructed image, etc. Once the convergence condition is met, an optimized transmittance map is obtained. Finally, edge-preserving filtering is applied to output an accurate target transmittance map. Specifically, a method for optimizing the initial transmittance map based on the hazy image to be processed and the global atmospheric light values ​​to obtain a target transmittance map may include: A1. Perform image dehazing processing based on the foggy image to be processed, the global atmospheric light value, and the initial transmittance map to obtain the initial restored image; A2. Construct a correction term based on the initial restored image, the global atmospheric light value, and the initial transmittance map to obtain the initial correction term; A3. Based on the initial correction term, the initial restored image, the initial transmittance map, and the global atmospheric light value, the initial transmittance map is iteratively updated to obtain a reference transmittance map; A4. If the reference transmittance map meets the preset convergence condition, the reference transmittance map is determined as the optimized transmittance map. A5. Perform edge-preserving filtering on the optimized transmittance map to obtain the target transmittance map.

[0035] The initial restored image can refer to the first dehazed image obtained by back-deriving from the atmospheric scattering model, using the hazy image to be processed, the global atmospheric light value, and the initial transmittance map as input. This initial restored image can be regarded as the basis for the subsequent construction of correction terms.

[0036] Specifically, the initial transmittance map can be used. and global atmospheric light value The initial restored image can be obtained using the following formula, i.e., the initial restored image estimate can be calculated. : ; in, It can refer to the initial transmittance. Estimated, pixels Located in color channel The initial restored (or defogging) intensity value; It can refer to the input of a foggy image at the pixel level. Color channel The intensity value; It can refer to global atmospheric light In color channels The components on; It can refer to a small positive number (such as 0.05) to prevent the denominator from being zero and to ensure numerical stability; Can be pointed to and The larger value in; It can refer to the color channel index, and the value is simply red ( ),green( ) and blue ( ).

[0037] Optionally, to simplify the calculation, the initial restored image can be estimated as described above. Convert to grayscale brightness The specific formula is as follows: in, It can refer to pixels The brightness value of the initial restored image can be obtained from the initial restored intensity values ​​of the RGB three channels, i.e. , and The values ​​are obtained by weighting according to human visual perception; the coefficients 0.299, 0.587 and 0.114 can be the standard luminance conversion coefficients for the R, G and B channels, respectively.

[0038] The correction term can refer to the error compensation factor calculated based on the initial restored image, global atmospheric light values, and the initial transmittance map. Its function is to correct the estimation deviations of the initial transmittance map in areas such as the sky and highlights. The initial correction term can refer to the correction term constructed during the first iteration. This initial correction term can be regarded as the initial basis for the iterative update of the transmittance map.

[0039] Specifically, construct correction terms, such as initial correction terms. The initial approximation can be obtained by referring to the following steps: in, It can refer to the initial correction term, that is, the term applied to pixels. The larger the value of the initial correction term, the better it represents the initial transmittance. The larger the correction magnitude; It can refer to the correction intensity coefficient, which is an adjustable hyperparameter that can be used to control the overall influence intensity of the correction term; It can refer to the pixels calculated in the aforementioned steps. The brightness value of the initial restored image; It can refer to the pixels calculated in the aforementioned steps. Initial transmittance at the location; It can refer to the global atmospheric light value.

[0040] Iterative updates can be understood as a process of continuously refining the transmittance map through iterative calculations, using initial correction terms, initial restored images, initial transmittance maps, and global atmospheric light values ​​as inputs. It's important to note that each iteration generates more accurate transmittance data. The reference transmittance map refers to the intermediate transmittance map obtained after each iteration. This reference transmittance map can be used to determine whether a preset convergence condition has been met; in other words, it can be considered the basis for determining convergence.

[0041] A preset convergence condition can refer to a threshold criterion for determining whether the iteration should stop. Optionally, this preset convergence condition can typically be that the average change in the reference transmittance map generated during two adjacent transmittance map update iterations is less than a preset average change threshold, or that the preset convergence condition is that the number of transmittance map update iterations reaches a preset maximum update iteration value. Here, the average change in the reference transmittance map between two adjacent iterations being less than the set threshold can be understood as the average change between the transmittance map in the current transmittance map update iteration and the transmittance map in the previous transmittance map update iteration being less than the preset average change threshold.

[0042] An optimized transmittance map refers to a transmittance map obtained after meeting preset convergence conditions, where the deviation has been significantly reduced. Edge-preserving filtering can be understood as an image filtering operation that smooths noise in the transmittance map while preserving the sharpness of object edges in the image, thus avoiding edge blurring after dehazing. The target transmittance map is the final transmittance map after the aforementioned edge-preserving filtering.

[0043] The iterative transmittance update process can mainly address deviations in overall brightness areas (such as the sky), but it may introduce or fail to eliminate local noise or unevenness. Guided filtering, on the other hand, smooths out subtle fluctuations within the transmittance map while maintaining sharp object edges, making the final transmittance map more physically plausible (transmittance should be relatively smooth locally), thereby further improving the visual quality of the restored image.

[0044] Specifically, in order to remove noise that may be introduced by iteration and preserve the edges, guided filtering can be used. Perform smoothing. Optional, return to the original hazy image. As a guide graph, the filter radius and regularization parameters can be set; for example, radius = 32. To obtain the final transmittance map with edge preservation. For the specific formula, please refer to the following: in, It can refer to the final transmittance map after being smoothed by guided filtering; This can guide the filtering function; here, the transmittance map after iterative optimization is used. The input image is the original foggy image. To guide the image, perform edge-preserving smoothing operations.

[0045] In this embodiment, the estimated bias of the initial transmittance map is effectively corrected through closed-loop processing of restoration, correction, iteration, and filtering. This solves the problem of halo artifacts and color distortion that are prone to occur in the sky and highlight areas by traditional dark channel prior methods. The iterative process adaptively optimizes the transmittance data without relying on complex image segmentation or deep learning training. Edge-preserving filtering further improves the accuracy of the transmittance map, ensuring the clarity of details and natural colors in the final dehazed image, thus balancing the robustness and practicality of the algorithm.

[0046] In one possible implementation, when iteratively updating the initial transmittance map, the initial transmittance map can first be corrected with an initial correction term to obtain an updated transmittance map. Then, the restored image is updated by combining the hazy image and atmospheric light values. Subsequently, a new correction term is constructed based on the updated restored image and transmittance map. Finally, the transmittance map is corrected again with the new correction term to obtain the reference transmittance map for this iteration. Specifically, a method for iteratively updating the initial transmittance map based on the initial correction term, the initial restored image, the initial transmittance map, and the global atmospheric light values ​​to obtain a reference transmittance map may include: B1. Perform transmittance update processing on the initial transmittance map according to the initial correction term to obtain an updated transmittance map; B2. Based on the foggy image to be processed, the global atmospheric light value, and the updated transmittance map, perform image update processing to obtain the updated restored image; B3. Based on the updated restored image, the global atmospheric light value, and the updated transmittance map, perform correction term update processing to obtain the updated correction term; B4. Update the updated transmittance map according to the updated correction terms to obtain a reference transmittance map.

[0047] As mentioned above, the initial correction term can refer to the error compensation factor constructed based on the initial restored image, global atmospheric light value, and initial transmittance map during the first iteration, used to correct the deviation of the initial transmittance map. Updating the transmittance map can be understood as the calculation process of compensating and correcting the pixel values ​​of the current transmittance map using the correction term. It should be noted that the purpose of updating the transmittance map is to reduce the estimated error of transmittance. The updated transmittance map can refer to the intermediate transmittance map obtained after correcting the initial transmittance map with the initial correction term.

[0048] Updating the restored image can be understood as a process based on an atmospheric scattering model, incorporating the hazy image to be processed, global atmospheric light values, and an updated transmittance map into calculations to generate a new restored image. The updated restored image can refer to a new dehazed image derived from the updated transmittance map. It is understood that the accuracy of the updated restored image can be superior to the initial restored image, and this updated restored image can be seen as the basis for constructing new correction terms.

[0049] The correction term update process can be understood as the process of recalculating the error compensation factor based on the updated restored image, global atmospheric light values, and updated transmittance map, thereby making the correction term more suitable for the current, more accurate image data. The updated correction term refers to the new error compensation factor obtained after the correction term update process, which can be used to further correct and update the transmittance map.

[0050] The reference transmittance map can refer to the result obtained after the updated transmittance map has been corrected again by the update correction term. It should be noted that there can be one or more reference transmittance maps, and this application does not impose any limitation on this. The reference transmittance map can be considered an intermediate product in the iterative update process and can be used to determine whether a preset convergence condition is met. If the reference transmittance map meets the preset convergence condition, the iterative update process ends; if the reference transmittance map does not meet the preset convergence condition, the iterative update process is repeated.

[0051] Specifically, the above iterative update process may include the following sub-steps: b101: Initialize the iterative update index, i.e. ,set up ; b102: Based on the current transmittance diagram The updated and restored image is obtained; the specific formula is as follows: ; in, It can refer to the restored (dehazed) image intensity (usually brightness or channel values) at pixel x estimated during the i-th iteration update. (x) can refer to the transmittance at pixel x in the i-th iteration; It can refer to the original foggy image, i.e., the foggy image to be processed; It can refer to the global atmospheric light value; It can be a small positive number (such as 0.05) to prevent the denominator from being zero, thus ensuring numerical stability; that is, to enhance numerical stability, the current transmittance in the denominator can be used. (x) with a small positive number Take the maximum value.

[0052] b103: Based on the current restored image and current transmittance Calculate the current correction term to obtain the updated correction term. The specific formula is as follows: in, It can refer to the transmittance correction term for pixel x in the i-th iteration; It can refer to the intensity of the restored image at pixel x estimated in the i-th iteration; It can refer to the transmittance at pixel x during the i-th iteration; It can refer to the correction strength coefficient, and experimental verification shows that the optimal value can be 0.8; It can refer to the global atmospheric light value.

[0053] b104: Update the transmittance map using the current correction term to obtain the reference transmittance map, as shown in the following formula: in, It can refer to the updated transmittance at pixel x in the (i+1)th iteration; It can refer to the initial transmittance calculated in the aforementioned steps, that is, the transmittance used as a reference. It can refer to the correction term calculated in the i-th iteration.

[0054] b105: Determine if the preset convergence condition is met; if so, stop the iterative update and output the result. As an optimized transmittance map; otherwise, let Then return to step b102 above to perform the next iteration update process.

[0055] Specifically, the preset convergence condition is that the average change of the reference transmittance map generated during two adjacent transmittance map update iterations is less than a preset average change threshold, or the preset convergence condition is that the number of transmittance map update iterations reaches a preset maximum update iteration value.

[0056] Optionally, the average change in the reference transmittance map generated during two adjacent transmittance map update iterations. The calculation process can be found in the following formula: in, It can refer to the average absolute change of the entire transmittance map after the i-th iteration, and can be used to measure the magnitude of the iterative update; It can refer to the total number of pixels in an image; It can refer to the pixel domain of an image, that is, the set of all pixel locations; It can refer to the transmittance at the i-th iteration; It can refer to the transmittance at the (i+1)th iteration.

[0057] In other words, if or Then the above iterative update loop can be terminated, and the final optimized transmittance map can be set as follows: Otherwise, it can be made This will continue to the next iteration. This can refer to a preset average change threshold. It can refer to the preset maximum number of iterations, that is, the preset maximum number of update iterations.

[0058] In this embodiment, an adaptive iterative optimization of the transmittance map is achieved through a closed loop of transmittance update, restored image update, correction term update, and re-correction of transmittance. This loop can reduce the estimation deviation of the initial transmittance map in complex areas such as the sky and highlights in each round, avoiding the artifact problem caused by traditional single estimation. The entire iterative process does not rely on additional image segmentation or training data. It can improve transmittance accuracy through its own data closed loop iteration, which greatly enhances the robustness of the algorithm and the stability of the dehazing effect.

[0059] In one possible implementation, this application also provides a method for optimizing the selection of hyperparameters, specifically involving the correction intensity coefficient. And the maximum number of iterations (i.e., the preset maximum number of update iterations). The optimization determination steps may specifically include the following steps: C1. Obtain a training image sample dataset, which includes foggy images and fog-free images corresponding to the foggy images; C2. Perform grid search processing on the training image sample dataset to obtain a set of hyperparameter combinations; C3. Based on each hyperparameter combination in the hyperparameter combination set, perform dehazing processing on each training image sample data in the training image sample dataset to obtain a set of dehazed result images; C4. Perform multi-index robust score evaluation on each dehazing result image in the set of dehazing result images to obtain a composite evaluation score set; C5. Based on each composite evaluation score in the composite evaluation score set, calculate the average composite evaluation score of the training image sample data corresponding to each hyperparameter combination to obtain the set of average composite evaluation scores. C6. Select the maximum average composite evaluation score from the set of average composite evaluation scores, and take the hyperparameter combination corresponding to the maximum average composite evaluation score as the optimal hyperparameter combination; the optimal hyperparameter combination includes the preset maximum update iteration value.

[0060] The training image sample dataset can refer to a dataset consisting of multiple paired samples. Each sample in this dataset can contain a foggy image taken on a foggy day, and a corresponding real, fog-free image (i.e., a fog-free reference image). This training image sample dataset can be considered as the basic test data for hyperparameter optimization.

[0061] Grid search processing can refer to a hyperparameter optimization method, which is the process of cross-combining candidate values ​​of different hyperparameters to generate all possible combinations of hyperparameters, given a preset range of candidate hyperparameters, such as a candidate set of iteration counts or a candidate interval of correction intensity coefficients.

[0062] The hyperparameter combination set can refer to a set containing all candidate hyperparameter pairs obtained through grid search. It is understood that each hyperparameter combination in this set corresponds to a set of hyperparameters to be tested. For example, hyperparameter combination 1 could be: 3 iterations + 0.8 correction strength coefficient. It should be noted that this embodiment uses a hyperparameter combination including iterations and correction strength coefficient as an example for illustration, and does not constitute a limitation on this application. Optionally, the hyperparameter combination may also include other hyperparameters, and this application does not impose any restrictions on this.

[0063] The set of dehazed result images can refer to the set of all dehazed images obtained by substituting each set of hyperparameters in the hyperparameter combination set into the dehazing method, thereby achieving the dehazing of all hazy images in the training image sample dataset.

[0064] Multi-metric robust score evaluation can be understood as an evaluation method based on the difference between the dehazed image and the corresponding real haze-free image. This method selects multiple objective evaluation metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Color Difference (ΔE), and then standardizes the scores of each metric using a robust Z-score (calculated based on the median and median absolute deviation of the metrics). This multi-metric robust score evaluation can reduce the interference of outliers on the evaluation results.

[0065] The composite evaluation score set refers to the set of comprehensive evaluation scores for each dehazed image obtained by fusing multiple robust index scores, which can comprehensively reflect the dehazing quality of a single image. The composite evaluation score average set refers to the set containing the average composite evaluation scores for all hyperparameter combinations. The composite evaluation score average can also refer to the arithmetic mean of the composite evaluation scores for a given set of hyperparameters across all training images, reflecting the average dehazing effect of that set of hyperparameters on the entire dataset.

[0066] The optimal hyperparameter combination can refer to the hyperparameter combination with the largest average value in the set of average composite evaluation scores. This optimal hyperparameter combination can enable the dehazing method to achieve the best overall effect on the training dataset, and may include the optimized preset maximum value of the update iteration.

[0067] Specifically, to ensure the best performance of this application in different scenarios, two core hyperparameters can be adjusted: the correction intensity coefficient k and the maximum number of iterations. Optimization is then performed. This application proposes a grid search strategy based on robust Z-score fusion of multiple indicators. Specific steps are as follows: c101: Defines the parameter search space. The correction intensity coefficient k can be taken in the interval [0.1, 1.0] with a step size of 0.1; the maximum number of iterations... Values ​​can be taken from the set {1, 2, 3, 4, 5}. For (k, Each combination of these methods can be used to run the dehazing method provided in this application on the training segmentation of the training image set, such as the Non-Homogeneous Hazy and Haze-Free Image Dehazing Benchmark (NH-HAZE).

[0068] c102: Calculate the evaluation score for a single image. For each training image i, given (k, Under the condition that the dehazing result is calculated relative to the real haze-free image, three standard indicators are used: PSNR (the higher the value, the better), SSIM (the higher the value, the better), and ΔE (CIEDE2000 color difference, the lower the value, the better).

[0069] Because these three metrics have different dimensions and numerical ranges, direct summation would be unfair. Therefore, this application introduces a robust Z-score to normalize each metric. That is, for a metric x (representing PSNR, SSIM, or ΔE), its values ​​across all training images form a set X, and the median of this set is calculated. and median absolute deviation For the specific formula, please refer to the following: in, Can refer to indicators The median absolute deviation of the set of values ​​across all training images can be used to measure the dispersion of the metric values ​​and is not sensitive to outliers. Can refer to indicators The median of the values ​​across all training images.

[0070] Furthermore, a robust standard deviation estimate can be calculated using a constant factor of 1.4826. The specific formula is as follows: in, Can refer to indicators A robust estimate of the standard deviation; the constant 1.4826 can be used to convert the median absolute deviation (MAD) into an approximately unbiased estimate of the standard deviation (assuming the data follows a Gaussian distribution).

[0071] Optionally, for the index value of image i Its robust Z-score The calculation formula can be found below: in, It can refer to the robust Z score of image i on index x, and may not represent the degree of deviation of the index value of the image from the median of all images (in robust standard deviation). It can refer to the value of image i on index x; Can refer to indicators Robust standard deviation estimation; Can refer to indicators The median of the values ​​across all training images.

[0072] Optionally, a composite Z-score can be constructed. To comprehensively evaluate the dehazing quality of image i, the specific formula is as follows: in, The overall quality score of image i can be obtained by adding the Z scores of PSNR and SSIM (the higher the better) and subtracting the Z score of ΔE (the lower the better); This can be a robust Z-score for image i on the PSNR metric; This can be a robust Z-score for image i on the SSIM metric; This can be a robust Z-score for image i on the ΔE metric. It is understandable that this... A larger value of z_comp(i) indicates better overall dehazing quality for image i. In other words, a larger z_comp(i) value indicates better overall dehazing performance.

[0073] c103: Determine the optimal parameter combination: For each group (k, This can be used to calculate the average composite Z-score of all training images. The specific formula is as follows: in, It can refer to a specific combination of parameters The average overall quality score of all N training images; It can refer to the total number of training images; The overall quality score of image i can be determined. It should be noted that it is possible to choose an image such that... The largest value (k, The combination of parameters is used as the optimal hyperparameter.

[0074] In this embodiment, the optimal parameters determined on the NH-HAZE training set by the above method can be: k=0.8. =2. It should be noted that this set of parameters will be used consistently in all subsequent tests and applications of this application.

[0075] Optional, such as Figure 2A As shown, Figure 2A A flowchart illustrating the overall defogging process is provided. For example... Figure 2A As shown, the process begins by taking a foggy image as input to estimate the global atmospheric light value A, and then calculating the initial transmittance using the dark channel prior (DCP). Next, the initial image is reconstructed using the foggy image, atmospheric light value, and initial transmittance. Subsequently, an iterative transmittance update stage is entered, and the transmittance is updated cyclically using a formula until the convergence condition is met. The final transmittance is then further optimized by guided filtering and substituted into the atmospheric scattering model formula to complete the final reconstruction, ultimately outputting a defogging image.

[0076] In this embodiment, by constructing a training sample dataset with and without fog, a grid search is used to traverse all possible combinations of hyperparameters. Combined with robust multi-index score evaluation and screening using the average of composite evaluation scores, the optimal hyperparameter combination can be objectively and comprehensively selected. The entire process avoids the subjectivity and limitations of manual parameter tuning, effectively improving the scientific rigor and reliability of hyperparameter optimization. At the same time, the introduction of robust Z-scores reduces the interference of outlier samples on the evaluation results, giving the selected optimal hyperparameter combination stronger generalization ability, enabling it to adapt to foggy images of different concentrations and scenarios, and significantly improving the overall robustness and practical value of the defogging method.

[0077] The method provided in this application significantly improves the overall performance and applicability of single-image dehazing by organically integrating dark channel priors with iterative transmittance correction logic. First, compared to traditional dark channel prior dehazing methods that rely solely on single transmittance estimation, leading to issues such as halo effects in sky areas and color distortion in highlight areas, this application constructs an iterative update closed loop of the initial restored image, correction terms, and transmittance map. This adaptively corrects the estimation deviation of the initial transmittance map in complex scenes (such as skies, strong light sources, and large areas of high brightness), effectively suppressing artifact generation in the dehazed image while accurately restoring the scene's detailed textures and natural colors, resulting in dehazing quality far exceeding traditional methods. Second, this application does not rely on complex preprocessing or additional dependencies such as image region segmentation or deep learning model training. Based solely on the core logic of the atmospheric scattering physics model and dark channel priors, combined with pixel-level iterative optimization, it effectively controls computational complexity while ensuring dehazing effects, achieving near real-time processing capabilities. This solves the pain point of traditional complex dehazing algorithms being difficult to deploy in resource-constrained scenarios such as embedded devices and mobile terminals. Furthermore, this application determines the optimal number of iterations and correction intensity coefficients through hyperparameter optimization, and selects parameters using a multi-index robust evaluation system. This enables the algorithm to exhibit strong robustness to foggy images under varying concentrations of fog, haze, and complex weather conditions, achieving adaptive defogging without manual parameter adjustment. Applicable scenarios cover a variety of practical needs, including autonomous driving, security monitoring, outdoor photography, and remote sensing imaging. Finally, the core logic of this application is derived from a physical model, avoiding the black-box problem of deep learning methods. The algorithm is highly interpretable and reliable, demonstrating significant advantages in industrial applications with stringent stability and security requirements, and showcasing outstanding practicality and applicability.

[0078] In one possible implementation, to objectively and comprehensively verify the effectiveness and advancement of this application (named E-DCP-ITU), four representative and widely cited algorithms in the field of physical model-based single-image dehazing were selected as baselines for comparison and compared under the same experimental environment. The specific details are as follows: 1. Comparison Method DCP: The classic dark channel prior method proposed by He et al. serves as the basis for the improvement of this invention.

[0079] CAP: The color attenuation prior method proposed by Zhu et al. is known for its high efficiency.

[0080] Haze-Lines: A nonlocal dehazing method based on fog line priors proposed by Berman et al., which excels at improving transmittance consistency.

[0081] Multi-DCP: The multi-scale dark channel prior fusion method proposed by Kim et al. represents the current advanced level of improving DCP to enhance color fidelity.

[0082] 2. Experimental Setup Datasets: The synthetic dataset RESIDE (SOTS-Outdoor) and the real dataset NH-HAZE were used.

[0083] Evaluation metrics: Peak signal-to-noise ratio (PSNR, the higher the better), structural similarity (SSIM, the higher the better), and CIEDE2000 color difference (ΔE, the lower the better).

[0084] Implementation details: All comparison algorithms are implemented strictly according to their original literature and run in the same experimental environment (hardware, software, parameter settings) to ensure the fairness of the comparison.

[0085] 3. Quantitative Results Analysis RESIDE dataset: As shown in Table 1, this invention achieves optimal values ​​in both PSNR (19.40 dB) and SSIM (0.8940), significantly outperforming DCP (16.35 dB, 0.8608), CAP (19.36 dB, 0.8673), Haze-Lines (13.80 dB, 0.7480), and Multi-DCP (18.53 dB, 0.6540). Regarding color fidelity (ΔE), the result is 8.19, very close to the optimal value of 7.85 (Multi-DCP).

[0086] NH-HAZE dataset: As shown in Table 2, in more challenging real-world scenarios, this invention achieves the best performance in three metrics: PSNR (13.18dB), SSIM (0.5132), and ΔE (20.67), comprehensively surpassing all comparison methods.

[0087] Table 1. Comparison of quantitative results of RESIDE (SOTS-Outdoor)

[0088] Table 2. Comparison of NH-HAZE Quantitative Results

[0089] 4. Qualitative Results Analysis Figure 2C and Figure 2D Visual contrast results on the RESIDE and NH-HAZE datasets are presented respectively. It can be observed that: DCP: Produces noticeable dark areas and halo artifacts in the sky and bright areas, with the overall color being darker.

[0090] CAP: Dehazing is not thorough enough, leaving a lot of haze in the image and insufficient detail recovery.

[0091] Haze-Lines: Uneven defogging results, with some areas being too dark and others still appearing foggy.

[0092] Multi-DCP: Color fidelity has been improved, but color distortion and localized over-darkness still exist in some areas.

[0093] This application (E-DCP-ITU) restores images with natural overall brightness, clear details, and high color fidelity. In particular, it effectively suppresses halo artifacts in areas where traditional DCP fails, such as the sky, achieving the most balanced and natural visual effect.

[0094] 5. Operational efficiency analysis Tables 3 and 4 show the average runtime (milliseconds) of each method on the RESIDE and NH-HAZE datasets, respectively. The average processing time of this invention is approximately 39.75 milliseconds on RESIDE and approximately 324.20 milliseconds on NH-HAZE. Although slightly slower than the lightest CAP and basic DCP, it is significantly faster than the computationally complex Haze-Lines (approximately 7-9 times slower) and Multi-DCP (approximately 3.5 times slower), achieving optimal dehazing quality while maintaining near real-time processing capabilities, meeting the efficiency requirements of practical applications.

[0095] Table 3. Comparison of RESIDE runtime (ms)

[0096] Table 4. Comparison of NH-HAZE running times (ms)

[0097] 6. Conclusion In summary, this invention effectively addresses the key shortcomings of traditional dark channel priors in bright and low-texture regions by introducing physically derived correction terms, iterative update mechanisms, and scientific parameter optimization methods. Extensive experimental results demonstrate that this invention outperforms current mainstream physical model dehazing methods in terms of dehazing quality (PSNR, SSIM, ΔE), visual effects, and operational efficiency. While maintaining algorithm interpretability and low computational complexity, it achieves a significant performance improvement, exhibiting high practical value and promising prospects for wider application.

[0098] For examples consistent with the above embodiments, please refer to... Figure 3 , Figure 3 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application, such as... Figure 3As shown, it includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions. The program includes instructions for performing the following steps. Acquire the foggy image to be processed; The atmospheric light value is estimated based on the foggy image to be processed to obtain the global atmospheric light value. Based on the dark channel prior, the transmittance map is calculated and processed according to the foggy image to be processed to obtain the initial transmittance map. Based on the foggy image to be processed and the global atmospheric light value, the initial transmittance map is optimized and iterated to obtain the target transmittance map; Based on the atmospheric scattering model, image dehazing is performed according to the foggy image to be processed, the global atmospheric light value, and the target transmittance map to obtain the target restored image.

[0099] The above mainly describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the terminal includes the corresponding hardware structure and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0100] This application embodiment can divide the terminal into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0101] For those consistent with the above, please refer to Figure 4 , Figure 4 This application provides a schematic diagram of a single-image dehazing device based on iterative transmittance correction, as illustrated in this embodiment. Figure 4 As shown, the device includes: Acquisition unit 101 is used to acquire the foggy image to be processed; The first processing unit 102 is used to perform atmospheric light value estimation processing based on the foggy image to be processed to obtain the global atmospheric light value. The second processing unit 103 is used to perform transmittance map calculation processing based on the dark channel prior and the foggy image to be processed to obtain an initial transmittance map. The third processing unit 104 is used to perform optimization and iterative processing on the initial transmittance map based on the foggy image to be processed and the global atmospheric light value to obtain the target transmittance map. The fourth processing unit 105 is used to perform image dehazing based on the atmospheric scattering model, according to the foggy image to be processed, the global atmospheric light value, and the target transmittance map, to obtain the target restored image.

[0102] In one possible implementation, the third processing unit 104 is configured to perform optimization iterative processing on the initial transmittance map based on the foggy image to be processed and the global atmospheric light value to obtain a target transmittance map, specifically for: Image dehazing is performed based on the foggy image to be processed, the global atmospheric light value, and the initial transmittance map to obtain the initial restored image; The initial correction term is constructed based on the initial restored image, the global atmospheric light value, and the initial transmittance map. Based on the initial correction term, the initial restored image, the initial transmittance map, and the global atmospheric light value, the initial transmittance map is iteratively updated to obtain a reference transmittance map; If the reference transmittance map satisfies the preset convergence condition, the reference transmittance map is determined as the optimized transmittance map. The optimized transmittance map is subjected to edge-preserving filtering to obtain the target transmittance map.

[0103] In one possible implementation, the third processing unit 104 is configured to iteratively update the initial transmittance map based on the initial correction term, the initial restored image, the initial transmittance map, and the global atmospheric light value to obtain a reference transmittance map, specifically for: The initial transmittance map is updated based on the initial correction term to obtain an updated transmittance map. Based on the foggy image to be processed, the global atmospheric light value, and the updated transmittance map, the restored image is updated to obtain the updated restored image. The updated correction terms are obtained by performing correction term update processing based on the updated restored image, the global atmospheric light value, and the updated transmittance map. The updated transmittance map is updated according to the updated correction terms to obtain a reference transmittance map.

[0104] In one possible implementation, the preset convergence condition is that the average change of the reference transmittance map generated during two adjacent transmittance map update iterations is less than a preset average change threshold, or the preset convergence condition is that the number of transmittance map update iterations reaches a preset maximum update iteration value.

[0105] In one possible implementation, the third processing unit 104 is further configured to: Obtain a training image sample dataset, which includes foggy images and fog-free images corresponding to the foggy images; A grid search process is performed on the training image sample dataset to obtain a set of hyperparameter combinations; Based on each hyperparameter combination in the hyperparameter combination set, dehazing is performed on each training image sample data in the training image sample dataset to obtain a set of dehazed result images; A multi-index robust score evaluation is performed on each dehazing result image in the set of dehazing result images to obtain a composite evaluation score set. Based on each composite evaluation score in the composite evaluation score set, calculate the average composite evaluation score of the training image sample data corresponding to each hyperparameter combination to obtain the set of average composite evaluation scores. The maximum average composite evaluation score in the set of average composite evaluation scores is selected, and the hyperparameter combination corresponding to the maximum average composite evaluation score is taken as the optimal hyperparameter combination; the optimal hyperparameter combination includes the preset maximum update iteration value.

[0106] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the single-image dehazing methods based on iterative transmittance correction described in the above method embodiments.

[0107] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the single-image dehazing methods based on iterative transmittance correction as described in the above method embodiments.

[0108] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0109] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0110] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0111] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0112] Furthermore, the functional units in the various embodiments of the application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software program module.

[0113] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0114] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.

[0115] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A single-image dehazing method based on iterative transmittance correction, characterized in that, The single-image dehazing method based on iterative transmittance correction includes: Acquire the foggy image to be processed; The atmospheric light value is estimated based on the foggy image to be processed to obtain the global atmospheric light value. Based on the dark channel prior, the transmittance map is calculated and processed according to the foggy image to be processed to obtain the initial transmittance map. Based on the foggy image to be processed and the global atmospheric light value, the initial transmittance map is optimized and iterated to obtain the target transmittance map; Based on the atmospheric scattering model, image dehazing is performed according to the foggy image to be processed, the global atmospheric light value, and the target transmittance map to obtain the target restored image.

2. The single-image dehazing method based on iterative transmittance correction according to claim 1, characterized in that, The step of optimizing and iteratively processing the initial transmittance map based on the foggy image to be processed and the global atmospheric light value to obtain the target transmittance map includes: Image dehazing is performed based on the foggy image to be processed, the global atmospheric light value, and the initial transmittance map to obtain the initial restored image; The initial correction term is constructed based on the initial restored image, the global atmospheric light value, and the initial transmittance map. Based on the initial correction term, the initial restored image, the initial transmittance map, and the global atmospheric light value, the initial transmittance map is iteratively updated to obtain a reference transmittance map; If the reference transmittance map satisfies the preset convergence condition, the reference transmittance map is determined as the optimized transmittance map. The optimized transmittance map is subjected to edge-preserving filtering to obtain the target transmittance map.

3. The single-image dehazing method based on iterative transmittance correction according to claim 2, characterized in that, The step of iteratively updating the initial transmittance map based on the initial correction term, the initial restored image, the initial transmittance map, and the global atmospheric light value to obtain a reference transmittance map includes: The initial transmittance map is updated based on the initial correction term to obtain an updated transmittance map. Based on the foggy image to be processed, the global atmospheric light value, and the updated transmittance map, the restored image is updated to obtain the updated restored image. The updated correction terms are obtained by performing correction term update processing based on the updated restored image, the global atmospheric light value, and the updated transmittance map. The updated transmittance map is updated according to the updated correction terms to obtain a reference transmittance map.

4. The single-image dehazing method based on iterative transmittance correction according to claim 2, characterized in that, The preset convergence condition is that the average change of the reference transmittance map generated during two adjacent transmittance map update iterations is less than a preset average change threshold, or the preset convergence condition is that the number of transmittance map update iterations reaches a preset maximum update iteration value.

5. The single-image dehazing method based on iterative transmittance correction according to claim 4, characterized in that, The method further includes: Obtain a training image sample dataset, which includes foggy images and fog-free images corresponding to the foggy images; A grid search process is performed on the training image sample dataset to obtain a set of hyperparameter combinations; Based on each hyperparameter combination in the hyperparameter combination set, dehazing is performed on each training image sample data in the training image sample dataset to obtain a set of dehazed result images; A multi-index robust score evaluation is performed on each dehazing result image in the set of dehazing result images to obtain a composite evaluation score set. Based on each composite evaluation score in the composite evaluation score set, calculate the average composite evaluation score of the training image sample data corresponding to each hyperparameter combination to obtain the set of average composite evaluation scores. The maximum average composite evaluation score in the set of average composite evaluation scores is selected, and the hyperparameter combination corresponding to the maximum average composite evaluation score is taken as the optimal hyperparameter combination; the optimal hyperparameter combination includes the preset maximum update iteration value.

6. A single-image dehazing device based on iterative transmittance correction, characterized in that, The device includes: The acquisition unit is used to acquire the foggy image to be processed; The first processing unit is used to perform atmospheric light value estimation processing on the foggy image to be processed to obtain the global atmospheric light value. The second processing unit is used to perform transmittance map calculation based on the dark channel prior and the foggy image to be processed to obtain an initial transmittance map. The third processing unit is used to perform optimization and iterative processing on the initial transmittance map based on the foggy image to be processed and the global atmospheric light value to obtain the target transmittance map. The fourth processing unit is used to perform image dehazing based on the atmospheric scattering model, the foggy image to be processed, the global atmospheric light value, and the target transmittance map to obtain the target restored image.

7. The single-image dehazing device based on iterative transmittance correction according to claim 6, characterized in that, The third processing unit is used to perform optimization and iterative processing on the initial transmittance map based on the foggy image to be processed and the global atmospheric light value to obtain the target transmittance map, specifically for: Image dehazing is performed based on the foggy image to be processed, the global atmospheric light value, and the initial transmittance map to obtain the initial restored image; The initial correction term is constructed based on the initial restored image, the global atmospheric light value, and the initial transmittance map. Based on the initial correction term, the initial restored image, the initial transmittance map, and the global atmospheric light value, the initial transmittance map is iteratively updated to obtain a reference transmittance map; If the reference transmittance map satisfies the preset convergence condition, the reference transmittance map is determined as the optimized transmittance map. The optimized transmittance map is subjected to edge-preserving filtering to obtain the target transmittance map.

8. The single-image dehazing device based on iterative transmittance correction according to claim 7, characterized in that, The third processing unit is used to iteratively update the initial transmittance map based on the initial correction term, the initial restored image, the initial transmittance map, and the global atmospheric light value to obtain a reference transmittance map, specifically for: The initial transmittance map is updated based on the initial correction term to obtain an updated transmittance map. Based on the foggy image to be processed, the global atmospheric light value, and the updated transmittance map, the restored image is updated to obtain the updated restored image. The updated correction terms are obtained by performing correction term update processing based on the updated restored image, the global atmospheric light value, and the updated transmittance map. The updated transmittance map is updated according to the updated correction terms to obtain a reference transmittance map.

9. A terminal, characterized in that, The device includes a processor, an input device, an output device, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the single-image dehazing method based on iterative transmittance correction as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the single-image dehazing method based on iterative transmittance correction as described in any one of claims 1-5.

Images