Unmanned aerial vehicle image defogging method based on global atmosphere optimization model

Abstract

The invention discloses an unmanned aerial vehicle image defogging method based on a global atmosphere optimization model, and the method comprises the steps: firstly designing a local window and dark channel optimization parameters to carry out the multi-channel pixel calculation, and extracting the dark channel information of an image through analyzing the darkest pixel value in the image; then, the dark channel information is used for estimating the transmissivity in the image, namely the light transmittance degree of particles in the atmosphere, and scattering and absorption interference of the particles to the light is reduced through guided filtering; in the process, the algorithm can adjust the transmissivity carefully so as to reflect the actual light propagation condition more accurately. And finally, selecting the brightest pixel point in the dark channel by the algorithm, and performing accurate estimation of the atmospheric light in combination with the transmissivity obtained by the previous optimization. Through the above steps, the algorithm can effectively remove the blurring and distortion of the image caused by haze, and recover a clear target image, thereby providing accurate unmanned aerial vehicle target information for a target recognition system.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision and image enhancement, and relates to a method for removing haze from UAV images based on a global atmospheric optimization model. Background Art

[0002] Under adverse weather conditions, such as meteorological factors like haze, smoke, and rainfall, images captured by image sensors often suffer from problems such as blurring and decreased contrast, which seriously affect the accuracy of target recognition and tracking. To overcome these visual obstacles, researchers have proposed various image dehazing and enhancement techniques in an attempt to restore the clarity and recognizability of images. Traditional image enhancement methods, such as histogram equalization and Laplace transform, can improve the contrast and detail information of images to a certain extent by adjusting the gray distribution of the images and enhancing edge information. These methods are widely used because of their simplicity and efficiency. However, they have limitations in dealing with complex natural environment changes. Especially in a thick fog or strong smoke environment, the loss of image details remains a difficult problem to solve. These traditional methods often rely on global image statistical information and lack consideration of local structures and scene depths. Therefore, when processing foggy images with highly non-uniform illumination and dynamic range, the effects are not ideal.

[0003] In recent years, the Dark Channel Prior (DCP) dehazing algorithm has become a research hotspot due to its excellent dehazing effect. The DCP algorithm is based on an observation that the dark channel of non-sky regions in natural images (i.e., the darkest value among the RGB channels of each pixel) tends to be very low. This prior knowledge enables the algorithm to inversely deduce the image in a haze-free scene, thereby effectively removing the haze effect. However, the DCP algorithm also has some drawbacks and limitations. First, the DCP algorithm performs poorly in processing sky regions because the dark channel of the sky region does not conform to the low-brightness prior, which may lead to color distortion in the dehazed sky region. Second, the DCP algorithm may have errors in estimating the image depth under thick fog conditions, resulting in over-enhancement or halo effects in the dehazed image.

[0004] The deficiencies of the above-mentioned existing technologies all affect the image processing capabilities of the target recognition system in complex environments and cannot meet its requirements for high-precision recognition and rapid response of UAV targets. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention proposes an unmanned aerial vehicle (UAV) image defogging method based on a global atmospheric optimization model. The algorithm design is simple and reasonable. The core of this method lies in constructing an accurate physical atmospheric light model, which can estimate the distribution and intensity of haze in the image, and then eliminate the influence of haze through inversion calculation to restore the true reflectivity of the target object.

[0006] To achieve the above object, the present invention adopts the following technical solutions: An unmanned aerial vehicle (UAV) image defogging method based on a global atmospheric optimization model, comprising the following steps:

[0007] Step 1: For the input blurred UAV image, extract the minimum value information from each of the red, green, and blue color channels and synthesize them to obtain a complete dark channel image;

[0008] Step 2: Design a local window to improve the efficiency of calculating the local minimum value of the dark channel, and introduce a low-brightness pixel enhancement parameter to make the calculation of the dark channel more accurate;

[0009] Step 3: Use the fog effect adjustment coefficient and the initial atmospheric light intensity to perform a preliminary transmittance estimation on the dark channel of the image. Then, refine the preliminary estimation value by introducing guided filtering to preserve the image edge information. Finally, adjust the weight coefficient and set the minimum transmittance threshold to ensure the accuracy of the transmittance estimation and the naturalness of the image;

[0010] Step 4: Screen out the set of the brightest pixel points from the dark channel image, use the pixel values of the corresponding original image of this set as the preliminary estimation of the atmospheric light. Then, correct the atmospheric light estimation value in combination with the optimized transmittance;

[0011] Step 5: Establish a reverse atmospheric scattering model based on the original image pixel values, transmittance, and atmospheric light intensity. Reverse and accurately calculate the pixel recovery value of the haze-free image by optimizing the transmittance and correcting the atmospheric light. At the same time, set the minimum transmittance threshold to avoid color distortion and effectively restore the original color and details of the image.

[0012] Compared with the prior art, the present invention has the following beneficial effects:

[0013] First, based on the traditional dark channel image, the present invention innovatively introduces dark channel optimization parameters and local windows, improving the calculation accuracy and efficiency of the dark channel. Then, the guided filtering technology is used for transmittance refinement to ensure the accuracy of transmittance estimation and the naturalness of the image. Finally, the optimized transmittance is used to correct the atmospheric light, and combined with the optimized transmittance and the corrected atmospheric light, the pixel recovery value of the image is calculated reversely, effectively restoring the original color and details of the image. The present invention has the characteristics of simplicity, high efficiency, and strong adaptability, realizing high-quality UAV image restoration in complex battlefield environments, not only retaining the clarity and details of the image but also significantly improving the accurate recognition ability of UAV targets. Description of the Drawings

[0014] Figure 1 is the implementation flowchart of the UAV image dehazing method based on the global atmospheric optimization model of the present invention:

[0015] Figure 2 is the schematic diagram of the atmospheric scattering model in the present invention; Detailed Embodiments

[0016] The following further describes the present invention in detail in conjunction with the drawings and specific embodiments.

[0017] As Figure 1 shown, the UAV image dehazing method based on the global atmospheric optimization model of the present invention includes the following steps:

[0018] Step 1: First, for the three basic color channels of the UAV input image - red (R), green (G), and blue (B), the minimum value maps are calculated respectively. Specifically, for each color channel c, by comparing the pixel values of each pixel point in the image and its neighborhood, the minimum pixel value in the channel is extracted, thereby constructing the corresponding minimum value map where, I c (y) is the pixel value of pixel y in channel c;

[0019] Subsequently, further comprehensive analysis is performed on these minimum value maps. For each pixel point x in the image, the minimum value is selected from the minimum value maps of the RGB three channels to generate the dark channel map J dark , and the dark channel map generation formula is as follows:

[0020] J dark (x) = min(M r (x), M g (x), M b (x))

[0021] Step 2: To improve the efficiency of calculating the local minimum of the dark channel and more effectively capture the dark information in the image, the present invention designs a local window Ω(x) = (2k + 1) × (2k + 1), where k is a parameter of the window radius, which determines the size of the local area. To further optimize the calculation of the dark channel, the present invention introduces a dark channel optimization parameter ρ. ρ is a constant greater than 1, which can enhance the effect of low-brightness pixels in the dark channel, so that the calculation of the dark channel is more sensitive to image degradation factors such as haze. Through this parameter, a more accurate dark channel calculation formula is obtained:

[0022]

[0023] Step 3: Use the fog effect adjustment coefficient ω and the initial atmospheric light intensity A to perform a preliminary transmittance estimation on the dark channel of the haze-free image. The formula for estimating the preliminary transmittance is:

[0024]

[0025] The preliminarily estimated transmittance map may be affected by noise and there is a certain loss in details. Therefore, the guided filtering technology is used to perform edge-preserving filtering on the preliminarily estimated transmittance to obtain the guided filtering result and introduce a weight coefficient α to adjust the balance between the preliminarily estimated transmittance and the guided filtering result . The specific optimization formula is as follows:

[0026]

[0027] To prevent the problem of image color distortion caused by too low transmittance, the present invention sets a minimum threshold t0 of the transmittance, and the value range is set between 0.1 and 0.2 to ensure that the transmittance will not be too low, so as to maintain the natural color of the image. The final transmittance estimation formula is:

[0028]

[0029] Step 4: Select the set of the brightest 0.1% pixel points {x1, x2,..., x dark} in the dark channel image J obtained in Step 2 as the area most affected by the atmospheric light. Then, among these selected pixel points, find the maximum pixel value at the corresponding position in the original image I to obtain a preliminary estimated value of the atmospheric light A n To further optimize the estimated result of the atmospheric light, the present invention designs an atmospheric light correction model, using the finally estimated transmittance t in Step 3 final ​(x) Further correct the atmospheric light A to obtain the corrected atmospheric light value Among them, Ω is the entire image area. The atmospheric light value obtained by such correction will not be too large or too small due to bright objects or noise in the image, and the color and contrast of the image are guaranteed.

[0030] Step 5: As Figure 2 As shown, during the process of atmospheric scattering imaging, the atmospheric light A, the foggy image I(x), and the clear image J(x) are coplanar and the lengths decrease in sequence. According to the atmospheric scattering model, the original image pixel value I(x) can be expressed as:

[0031] I(x) = J(x).t(x) + A.(1 - t(x))

[0032] Among them, J(x) is the pixel value of the fog-free image to be restored, t(x) is the transmittance, and A is the atmospheric light intensity. In order to restore J(x), based on the atmospheric scattering model, use the optimized transmittance t final (x) and the corrected atmospheric light A in Step 4 refined to calculate the pixel estimate value of the fog-free image To avoid the situation of dividing by zero, usually set a minimum transmittance t final (x) threshold (such as 0.1) during the calculation to avoid color distortion in some areas of the image due to too low transmittance.

[0033] The above description is only a specific example of the present invention and does not constitute any limitation to the present invention. Obviously, for professionals in the field, after understanding the content and principle of the present invention, various corrections and changes in form and details may be made without departing from the principle and structure of the present invention. However, these corrections and changes based on the idea of the present invention are still within the protection scope of the claims of the present invention.

Claims

1. An unmanned aerial vehicle image dehazing method based on a global atmospheric optimization model, characterized in that, It includes the following steps: Step 1: First, calculate the minimum value maps for the three basic color channels of the input image of the drone - red R, green G, and blue B respectively; specifically, for each color channel c, by comparing the pixel values of each pixel point in the image and its neighboring pixels, the minimum pixel value within the channel is extracted, thereby constructing the corresponding minimum value map where I c (y) is the pixel value of pixel y in channel c; Subsequently, further comprehensive analysis is carried out on these minimum value maps. For each pixel point x in the image, the smallest value is selected from the minimum value maps of the RGB three channels to generate the dark channel map J dark , and the generation formula of the dark channel map is as follows: J dark J(x) = min(M r (x), M g (x), M b (x)) Step 2: To improve the efficiency of calculating the local minimum value of the dark channel and more effectively capture the dark information in the image, the present invention designs a local window Ω(x)=(2k + 1)×(2k + 1), where k is a parameter of the window radius, which determines the size of the local area; to further optimize the calculation of the dark channel, a dark channel optimization parameter ρ is introduced. ρ is a constant greater than 1, which can enhance the role of low-brightness pixels in the dark channel, so that the calculation of the dark channel is more sensitive to the degradation factors of the haze image; through this parameter, a more accurate dark channel calculation formula is obtained: Step 3: Use the fog effect adjustment coefficient ω and the initial atmospheric light intensity A to perform a preliminary transmittance estimation on the dark channel of the fog-free image. The estimation formula for the preliminary transmittance is as follows: The initially estimated transmittance map may be affected by noise and there is a certain loss in details. Therefore, the guided filtering technique is used to perform edge-preserving filtering on the initially estimated transmittance to obtain the guided filtering result And a weight coefficient α is introduced to adjust the initially estimated transmittance and the guided filtering result to balance between them. The specific optimization formula is as follows: To prevent the problem of image color distortion caused by too low transmittance, a minimum threshold t0 of the transmittance is set, and the value range is set between 0.1 and 0.2 to ensure that the transmittance will not be too low, so as to maintain the natural color of the image; the final transmittance estimation formula is: Step 4: The dark channel image J obtained from Step 2 dark select the set of the brightest 0.1% pixel points {x1, x2,..., x n} as the region most affected by the atmospheric light. Then, among these selected pixel points, find the maximum pixel value at the corresponding position in the original image I to obtain a preliminary estimated value of the atmospheric light A To further optimize the estimated result of the atmospheric light, the present invention designs an atmospheric light correction model, and uses the transmittance t final (x) finally estimated in Step 3 to further correct the atmospheric light A to obtain a corrected atmospheric light value where Ω is the entire image region; the atmospheric light value obtained by such correction will not be made too large or too small due to bright objects or noise in the image, and the color and contrast of the image are guaranteed; Step 5: In the process of atmospheric scattering imaging, the atmospheric light A, the hazy image I(x) and the clear image J(x) are coplanar and their lengths decrease in turn; according to the atmospheric scattering model, the original image pixel value I(x) can be expressed as: I(x)=J(x)·t(x)+A·(1 - t(x)) Among them, J(x) is the pixel value of the haze-free image to be restored, t(x) is the transmittance, and A is the atmospheric light intensity; in order to restore J(x), based on the atmospheric scattering model, the transmittance t final (x) optimized in step 3 and the atmospheric light A corrected in step 4 refined are used to calculate the pixel estimate of the haze-free image To avoid division by zero, by setting a minimum transmittance t final (x) threshold, such as 0.1, to avoid color distortion in some regions of the image due to too low transmittance.

Images