A low-light image enhancement method, device, equipment and medium

Abstract

The application discloses a low-light image enhancement method and device, equipment and medium, and relates to the technical field of image processing. The application first introduces an adaptive brightness partition mechanism to determine an optimal intensity threshold, and divides the image domain into a bright area and a dark area based on the optimal intensity threshold, and then can only perform brightness enhancement on the dark area. When performing brightness enhancement on the dark area, a contrast enhancement model including a gradient consistency constraint term, a regularization term and a characteristic function is constructed. In the contrast enhancement model, the gradient consistency constraint term is subtracted by the gradient of an adaptive gradient adjustment function, and the regularization term is subtracted by an adaptive brightness adjustment function. The gradient of the adaptive gradient adjustment function can adaptively adjust the amplification of the contrast, and the adaptive brightness adjustment function can adaptively adjust the amplification coefficient of the brightness, so that the brightness of the dark area in the low-light image can be adaptively adjusted, and the brightness of the low-light image can be accurately enhanced.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method, apparatus, device, and medium for enhancing low-light images. Background Technology

[0002] Contrast enhancement in dark or overly bright images is currently a hot research topic. Low contrast not only leads to unclear or unpleasant images for the human eye, but also reduces the accuracy of automated visual inspections. Therefore, image contrast optimization has become a key step in image preprocessing. The main goal of image tone adjustment is to make hidden image features and details clearer, thereby improving image quality. At the same time, image tone adjustment is the foundation for geospatial imaging (land mapping, environmental monitoring), clinical diagnosis (functional MRI, tomographic reconstruction), and nanomaterial analysis (cryo-electron microscopy, STEM-EELS spectroscopy).

[0003] Image contrast enhancement methods can be broadly categorized into explicit and implicit methods, which can be further divided into three distinct categories: histogram methods, spectral methods, and spatial methods. Histogram methods enhance the dynamic range of the resulting image by redistributing or expanding numerical values. Histogram equalization (HE) is the most representative histogram method, using the cumulative distribution function of the input pixel intensity as the transformation function to obtain 15 uniformly distributed histograms. However, this method sometimes over-enhances the image, resulting in artifacts. Spectral methods primarily employ Fourier transform and wavelet transform as their 25 basic frameworks. They typically decompose the input low-light image into multiple scales or sub-bands, strategically applying nonlinear operations to the transform coefficients derived from these scales or sub-bands. However, spectral methods are significantly affected by lighting conditions, often failing to accurately distinguish between bright and dark areas, leading to significant differences in the processed image. Spatial methods directly modify image intensity based on certain spatial assumptions about the human visual system to enhance the image. However, spatial methods are influenced by human subjectivity, making accurate image enhancement difficult.

[0004] To address the problems and limitations of histogram, spectral, and spatial methods, researchers proposed a variational model incorporating gradient consistency constraints and regularization terms. The gradient consistency constraint involves subtracting the gradient of the input image from the gradient of the desired image, squaring the result, and then integrating it over the image domain. The regularization term involves subtracting a fixed constant from the desired image, squaring the result, and then integrating it over the image domain. However, because the subtracted gradient of the input image and the fixed constant are both fixed values, this variational model represents a fixed enhancement strategy. For low-light images with varying colors, this fixed enhancement strategy is insufficient to achieve accurate enhancement. Summary of the Invention

[0005] The embodiment of the present application provides a low-light image enhancement method, device, equipment and medium, which can solve the problem that in the prior art, the gradient consistency constraint term and the regularization term in the current variational model are subtracted from the gradient of the input image and a fixed constant, which is a fixed enhancement strategy, and the fixed enhancement strategy is difficult to achieve accurate enhancement for low-light images with different colors.

[0006] The embodiment of the present application provides a low-light image enhancement method, device, equipment and medium, which can solve the problem that in the prior art, the gradient consistency constraint term and the regularization term in the current variational model are subtracted from the gradient of the input image and a fixed constant, which is a fixed enhancement strategy, and the fixed enhancement strategy is difficult to achieve accurate enhancement for low-light images with different colors.

[0007] The embodiment of the present application provides a low-light image enhancement method, device, equipment and medium, which can solve the problem that in the prior art, the gradient consistency constraint term and the regularization term in the current variational model are subtracted from the gradient of the input image and a fixed constant, which is a fixed enhancement strategy, and the fixed enhancement strategy is difficult to achieve accurate enhancement for low-light images with different colors.

[0008] The embodiment of the present application provides a low-light image enhancement method, device, equipment and medium, which can solve the problem that in the prior art, the gradient consistency constraint term and the regularization term in the current variational model are subtracted from the gradient of the input image and a fixed constant, which is a fixed enhancement strategy, and the fixed enhancement strategy is difficult to achieve accurate enhancement for low-light images with different colors.

[0009] The embodiment of the present application provides a low-light image enhancement method, device, equipment and medium, which can solve the problem that in the prior art, the gradient consistency constraint term and the regularization term in the current variational model are subtracted from the gradient of the input image and a fixed constant, which is a fixed enhancement strategy, and the fixed enhancement strategy is difficult to achieve accurate enhancement for low-light images with different colors.

[0010] ;

[0011] Wherein: The gradient consistency constraint term is represented by The gradient of the output image The gradient of the adaptive gradient adjustment function is represented by Ω represents the image domain. The regularization term is represented by The balance parameter is represented by The adaptive brightness adjustment function is represented by The feature function is represented by The gradient consistency constraint term is used to adjust the differential structure of each pixel in the dark area including edge features and texture features, so as to adjust the contrast of each pixel in the dark area. The regularization term is used to enhance the brightness of each pixel in the dark area. The feature function is used to constrain the pixel value of each pixel in the dark area.

[0012] Preferably, when the low-light image is a gray image, the image domain of the gray image is divided into a dark area and a bright area, comprising:

[0013] The brightness value of each pixel in the gray image is obtained, and the average brightness value of all pixels in the gray image is calculated ;

[0014] The average brightness value A luminance threshold of the grayscale image is set, and the luminance threshold is 1.1 times of the average luminance value The image domain of the grayscale image is divided into a dark region and a bright region according to the luminance threshold.

[0015] The bright region in the grayscale image is a region with a luminance threshold higher than the luminance threshold, and the dark region in the grayscale image is a region with a luminance threshold lower than the luminance threshold.

[0016] Preferably, when the low-light image is a color image, the image domain of the color image is divided into a dark region and a bright region, comprising:

[0017] Each pixel in the color image is obtained, and each pixel is decomposed into three-channel components 、 and The maximum channel luminance value of each pixel is obtained, which is represented as:

[0018] ;

[0019] The global average maximum luminance value of all pixels is calculated, which is represented as:

[0020] ;

[0021] Wherein: x ∈Ω;

[0022] The luminance threshold of the color image is set according to the global average maximum luminance value The luminance threshold is 1.1 times of the global average maximum luminance value The image domain of the color image is divided into a dark region and a bright region according to the luminance threshold.

[0023] The bright region in the color image is a region with a luminance threshold higher than the luminance threshold, and the dark region in the color image is a region with a luminance threshold lower than the luminance threshold.

[0024] Preferably, the setting of the luminance threshold of the color image comprises:

[0025] The maximum channel luminance value is proportional to the three channels represented by hue, saturation and luminance HSV, and is normalized as:

[0026] ;

[0027] Then V is consistent with the maximum channel luminance value in the division.

[0028] Preferably, the adaptive gradient adjustment function in the contrast enhancement model Represented as:

[0029] ;

[0030] The adaptive brightness adjustment function within the contrast enhancement model Represented as:

[0031] ;

[0032] in: α Indicates the contrast parameter; β Indicates brightness parameter; Represents the bright areas of the image domain in a low-light image; This represents the dark area in the image domain of a low-light image;

[0033] The characteristic function Represented as:

[0034] ;

[0035] in: S This represents a closed bounded interval.

[0036] This invention also provides a low-light image enhancement device, comprising:

[0037] The image partitioning module is used to acquire low-light images and the brightness value of each pixel in the low-light images, and to calculate the average brightness value of all pixels in the low-light images.

[0038] A brightness threshold is set based on the average brightness value, and the image domain of the low-light image is divided into dark areas and bright areas based on the brightness threshold; where bright areas are areas with brightness values ​​higher than the brightness threshold, and dark areas are areas with brightness values ​​lower than the brightness threshold.

[0039] The image enhancement module is used to construct a contrast enhancement model, and based on the contrast enhancement model, enhance the brightness of each pixel in the dark area to obtain an image with enhanced brightness; the contrast enhancement model is represented as:

[0040] ;

[0041] in: This represents the gradient consistency constraint term. Indicates the output image gradient, Let Ω represent the gradient of the adaptive gradient adjustment function, and let Ω represent the image domain. Represents the regularization term. Represents the balance parameters. This represents the adaptive brightness adjustment function; The feature function is used for adjusting the pixel value of each pixel in the dark area.

[0042] The embodiment of the present application further provides an electronic device, comprising a memory and a processor.

[0043] The memory is used for storing a computer program.

[0044] The processor is used for executing the computer program stored in the memory, so that the steps of the low-light image enhancement method are realized.

[0045] The embodiment of the present application further provides a computer readable storage medium, used for storing a computer program, wherein the computer program is executed by a processor to realize the steps of the low-light image enhancement method.

[0046] The embodiment of the present application provides a low-light image enhancement method, device, equipment and medium, and the beneficial effects are as follows compared with the prior art.

[0047] The embodiment of the present application first introduces an adaptive brightness partition mechanism to determine an optimal intensity threshold, and simultaneously determines the optimal intensity threshold as 1.1 times of the average brightness value of the image domain, divides the image domain into a bright area and a dark area based on the optimal intensity threshold, and then only performs brightness enhancement on the dark area; and then constructs a contrast enhancement model comprising a gradient consistency constraint term, a regularization term and a feature function when performing brightness enhancement on the dark area, wherein the gradient consistency constraint term in the contrast enhancement model is the gradient of the expected image minus the gradient of the adaptive gradient adjustment function, and is squared and then integrated on the image domain, the regularization term is the expected image minus the adaptive brightness adjustment function, and is squared and then integrated on the image domain, and the feature function is further added to constrain the pixel value of each pixel in the dark area, that is, in the contrast enhancement model of the present application, the gradient of the adaptive gradient adjustment function is subtracted from the gradient consistency constraint term, and the adaptive brightness adjustment function is subtracted from the regularization term, the gradient of the adaptive gradient adjustment function can adaptively adjust the amplification of the contrast, and the adaptive brightness adjustment function can adaptively adjust the amplification coefficient of the brightness, so that the brightness of the dark area in the low-light image can be adaptively adjusted, and the brightness of the low-light image can be accurately enhanced. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 The embodiment of the present application provides a low-light image enhancement method, and a whole flowchart of the low-light image enhancement method is shown in the figure.

[0049] Figure 2A domain division result comparison schematic diagram of a low-light image enhancement method provided by an embodiment of the present application; (a), (d), (g) and (j) in the figure are test images; (b), (e), (h) and (k) in the figure are images obtained by using 0.299 +0.587 +0.114 A domain division result of a calculated gray image; (c), (f), (i) and (l) in the figure are images obtained by using the maximum image division method proposed by the present application to divide the region.

[0050] Figure 3 An image effect schematic diagram after setting a contrast parameter of a low-light image enhancement method provided by an embodiment of the present application;

[0051] Figure 4 A scene one enhancement effect comparison schematic diagram of an enhancement method proposed by the present application and other enhancement methods of a low-light image enhancement method provided by an embodiment of the present application;

[0052] Figure 5 A scene two enhancement effect comparison schematic diagram of an enhancement method proposed by the present application and other enhancement methods of a low-light image enhancement method provided by an embodiment of the present application;

[0053] Figure 6 A scene three enhancement effect comparison schematic diagram of an enhancement method proposed by the present application and a learning-based enhancement method of a low-light image enhancement method provided by an embodiment of the present application;

[0054] Figure 7 A scene four enhancement effect comparison schematic diagram of an enhancement method proposed by the present application and a learning-based enhancement method of a low-light image enhancement method provided by an embodiment of the present application;

[0055] Figure 8 A comparison schematic diagram of an original image and an enhancement effect after using an enhancement method proposed by the present application of a low-light image enhancement method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the above objectives, features and advantages of the present application more apparent and understandable, the specific embodiments of the present application are described in detail below with reference to the accompanying drawings. In the following description, a lot of specific details are set forth in order to facilitate a full understanding of the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the concept of the present application, so the present application is not limited to the specific embodiments disclosed below.

[0057] Reference is made to Figure 1The embodiment of the present application provides a low-light image enhancement method, aiming at the limitations existing in the histogram method, the spectrum method and the space method at the present stage, a simple variational model is proposed by the present researchers, which is expressed as:

[0058] (1)

[0059] Where: ∇ represents a first-order differential operator; Ω represents an image domain; g And f represent an input image and an expected image respectively; represents the average value of f in Ω.

[0060] Integral expression As a gradient consistency constraint, it ensures that f the differential structure of the original image is preserved.

[0061] Taking as a regularization term can reduce the variance of f , thereby eliminating the influence of non-uniform illumination.

[0062] At the same time, µ >0 is a parameter that balances variance reduction and detail preservation; when µ is small enough ( µ →0), the required image f is coincident with the input image g ; on the contrary, when µ is large enough, the expected image f is infinitely close to a constant; when µ becomes larger, the above-described variational model can be used as a high-pass filter.

[0063] Although the above-described variational model achieves partial satisfactory computing performance on irregular intensity distribution, it has disadvantages in parameter sensitivity and computing efficiency; for example, in the variational model (1), the selection of the screening coefficient µ has a significant influence on the enhancement effect, and its core is only based on empirical parameter adjustment or simple local gradient statistics; such a strategy may lead to detail loss or noise amplification in complex scenes (such as low-light noisy images), and lacks robust theoretical guidance; for high-resolution images (such as 4K or medical body data), the computing complexity significantly increases, and it is difficult to meet the real-time processing demand (such as mobile terminal or video stream scene).

[0064] In view of the problems existing in the above-described variational model, the present application proposes an innovative adaptive framework, which aims to optimize low-brightness images, specifically:

[0065] I. Enhancing gray-scale images.

[0066] In the variational framework, the low-light image contrast enhancement problem can be represented as: given an original low-light (low-contrast) image g The overall goal of the enhancement is to identify a map with... g Images with similar gradients but reduced variance f To compensate for uneven light, this invention first designs a contrast enhancement model, expressed as:

[0067] (2)

[0068] As can be seen from the above description, variational model (2) is sparser than variational model (1); on the one hand, the enhanced image obtained by variational model (1) is smoother, while the enhanced image obtained by variational model (2) tends to preserve the image structure; on the other hand, variational model (1) (i.e., |∇) f −∇ g |²) Based on the assumption ∇ f −∇ g It follows a Gaussian distribution, while the variational model (2) (i.e., |∇) f −∇ g |) then based on the assumption ∇ f −∇ g It conforms to the Laplace distribution; experiments show that ∇ f −∇ g It is closer to the Laplace distribution than the Gaussian distribution, therefore, variational model (2) is more reasonable and effective than variational model (1).

[0069] However, the critical condition of variational model (2) is highly nonlinear, which makes it very difficult to solve. In order to make variational model (2) simpler and easier to solve, this invention transforms variational model (2) as follows:

[0070] (3)

[0071] in: g This represents the average value over Ω, i.e. It replaced the variational model (2) used .

[0072] Although variational model (3) is simple and efficient, it requires the desired image. f Near pixel-independent average value This isn't entirely unreasonable for image enhancement, as it aligns with the requirement ∇ f Approaching ∇ g and limiting parameters µ Very small contradiction.

[0073] To achieve better low-light image quality, this invention will input the image... g and constant In the variational model (3), replace with two adaptive, pixel-dependent functions. u and h Therefore, the variational model (3) can be rewritten as:

[0074] (4)

[0075] Adaptive function u , h and characteristic function Represented as:

[0076] (5)

[0077] Contrast parameter α >1.

[0078] (6)

[0079] Among the brightness parameters β >0.

[0080] (7)

[0081] in: and Let represent the bright and dark parts of the image domain Ω, respectively; the second term in the variational model (4) serves not only as a regularization component but also, in part, as a data fidelity component; this is because of the adjustable function of the bright part in formula (6). h The value is defined as the value of the original image g(x).

[0082] To make the adaptive function u Differentiable in variational model (4), this invention redefines the adaptive function. u The format is as follows:

[0083] (8)

[0084] Where: G∗ represents Gaussian convolution; indicator function If the indicator function holds true when x∈Ξ, otherwise .

[0085] This invention will input image g The image domain Ω of a low-light image (i.e., an image with non-uniform lighting) is divided into bright areas. and Anbu , making ∩ =∅andΩ= ∪ Meanwhile, the present invention selects 1.1 ( Using the average intensity of g as the segmentation threshold, this invention selects 1.1. Instead The threshold for segmentation is chosen because the input image of this invention is a low-light image; specifically, if the following is selected... As the segmentation threshold, the interval [ 1.1 ] will also be marked as a bright area; however, since the input image is a low-light image, [ 1.1 Images within the interval are still relatively dark to the human visual system; for x∈ Bright elements should be preserved by ensuring that the desired image f closely matches the input image g(x); for Dark elements should be minimized by making the desired image f close to... To enhance this, a higher β can more clearly reveal occluded image features and details; additionally, the contrast parameter... Used to enhance dark areas The local gradient.

[0086] Characteristic function exist f The value of f is zero when it exceeds the closed bounded interval S, and positive infinity otherwise; in order to restrict f to the closed bounded interval S, It is added to the energy functional; in fact, two reasonable choices for S are [0,255] and [inff(x), sup f(x)]; for convenience, this invention will use S = [0,255].

[0087] II. Enhance color images.

[0088] In the variational model (4), the dynamic function u and h This invention is specifically developed for processing grayscale images under low-light conditions; it extends previous domain partitioning to include color images; dark areas refer to those pixels that are blurred, invisible, and poorly lit (e.g., Figure 2 (a) refers to the ceiling of the house, while the bright areas refer to those pixels that are clearly visible and well-defined within Ω (such as...). Figure 2 (a) Light and ground); The objective of this invention is to divide the pixels in Ω into two categories (dark areas and bright areas) and enhance the dark areas.

[0089] Based on this, the present invention first defines the largest image as follows:

[0090] (9)

[0091] in:( The ) indicates that the max operator is executed point by point on x∈Ω.

[0092] Then the image domain Ω is divided into the following two distinct parts:

[0093] (10)

[0094] (11)

[0095] in: .

[0096] like Figure 2 (c), (f), (i), and (l) in the figure respectively show the domain segmentation results of the low-light image calculated according to formulas (10)-(11), that is, for Figure 2 (a), (d), (g) and (j) are the domain segmentation results of the low-light image calculated according to formulas (10)-(11); Figure 2 In the equations (b), (e), (h), and (k), 0.299 is used for (a), (d), (g), and (j). +0.587 +0.114 The result of domain partitioning of the calculated grayscale image.

[0097] The research in this invention found that using the maximum image as a classifier can produce reasonable divisions because the maximum image value is proportional to the channel represented by hue, saturation, and brightness (HSV), where... ;therefore, V achieves the same correct split; conversely, other classifiers may lead to incorrect splits; such as... Figure 2 As shown in (b), (e), (h), and (k), it can be seen that some pixels are misclassified. For example, in (e), the window is incorrectly classified as a dark area, and in (h), the red life jacket is incorrectly classified as a dark area.

[0098] To address this problem, channel domain partitioning is one possible approach, but this method often leads to hue preservation issues; for example, for an element... ∈Ω, where the color intensity is determined by Give; at this time It is a relatively large value; if used (Or V channel) as the classifier, then Will be assigned to As a bright element; therefore As for the bright element, it will be kept at (220, 15, 15) as much as possible, while the hue will be... On the one hand, the domain is preserved; on the other hand, if the domain is divided according to the channel, It will be classified as in channel R Classified as in channels G and B ; therefore, will be kept around 220 as much as possible, while and will be enhanced; as a result, the hue at the point will not be preserved, which will lead to unnatural images; therefore, the channel domain division is an unreasonable method.

[0099] Based on this, the present application extends the variational model (4) to:

[0100] (12)

[0101] where and are expressed as:

[0102] (13)

[0103] (14)

[0104] where: ; .

[0105] The core of the present application is to divide the image domain into darker and brighter regions by applying an optimal threshold; then, further use appropriate terms to represent the dark part, so as to enhance the pixels in the dark region by adjusting the contrast and brightness parameters. The present application determines an optimal intensity threshold, which divides the image into different dark and bright regions within its algorithm framework; then, the model assigns different optimization standards to the segmented regions: the relatively dark region is intensity modulated by the calibrated contrast amplification coefficient and the brightness coefficient, while the high brightness region retains its inherent radiation characteristics with minimal deviation.

[0106] In the experiment, the present application sets the contrast parameter ; the reason is that increasing the value can enhance the brightness of dark pixels, but at the same time, it will also make their intensity level flat; a larger value is more preferred to strengthen the image structure and avoid the over-flattened enhancement result; when , the effect is best; and The model parameters of , The number of iterations is 1269, The number of iterations is 833, 862 and 860 respectively, .

[0107] As Figure 3 shown, Figure 3 (a) in (b) represents the original image, (b) represents the Enhanced image, (c) represents the image enhanced by , =0.7 Enhanced image, (d) represents the image enhanced by , =1 Enhanced image, (e) represents the image enhanced by , =1.3 Enhanced image, it can be seen that using will make the sky dark, which is an undesirable effect; in contrast, using helps to maintain the original color in the low-light image while restoring the trees and stone pillars; as increases, the trees and stone pillars become brighter and brighter.

[0108] Figures Figure 4 and Figure 5 show the comparison of the enhancement effect between the method of the present application and different enhancement methods; in Figure 4 , the sky area is shown, which cannot be accurately processed by LIME and RRM, and some details in the enhancement results of the NPE and RRM methods have been lost; in the outer wall area in Figure 4 , the enhancement effect of the present application achieves a more optimal visual effect compared with the NPE, LIME and RRM methods (for example, in the enlarged part of Figure 4 , the enhancement effect of the present application retains more details and has higher naturalness).

[0109] In Figure 5 , there is a phenomenon of detail loss in the results of NPE and RRM, such as the texture area between the bricks in Figure 5 ; at the same time, over-enhancement and color distortion are also observed in the results of LIME; compared with these methods, the method of the present application not only retains the color characteristics of the image itself when enhancing the low-light image (i.e. produces a more natural result), but also retains the texture details in the image.

[0110] As shown in Figure 6 and Figure 7 , the method of the present application is compared with other learning-based low-light image enhancement methods, including KinD++

[51] , RUAS

[50] and URetinex-Net

[52] ; KinD++ and RUAS adopt 330 post-processing methods to eliminate noise, which sometimes leads to problems such as loss of details, blurring and even degradation of image quality, in contrast, URetinex-Net aims to suppress noise and retain details, effectively improving the visibility of low-light images; however, its enhancement results often lead to color distortion and other phenomena; for example, Figure 6 the color of the leaves in (d) and Figure 7The sky color in (d). In contrast, the method of the present application performs better in color correction and noise suppression.

[0111] The present application uses four indicators (PSNR, SSIM, MAE and LPIPS) to evaluate the image quality to judge the enhancement effect; higher PSNR and SSIM values indicate better image quality; on the contrary, smaller MAE and LPIPS values indicate higher image quality.

[0112] As shown in Table 1, the quantitative results of the 100 images in the data set are shown; the results show that the method of the present application is generally better than other technologies in multiple indicators, but the performance on SSIM is slightly worse than Urentintx Net, and the performance on MAE is slightly worse than KinD++ and Urentintx Net; the performance of the method of the present application on PSNR and LPIPS is significantly better than all other methods, proving the effectiveness of the proposed method; in addition, the method of the present application obtains the highest average ranking, which is obtained by aggregating the ranking of each indicator.

[0113] Table 1 Quantitative results of 100 images on different indicators

[0114]

[0115] As Figure 8 shown, for the comparison of the original image and the enhanced effect after using the enhancement method of the present application, it can be seen that the enhancement method of the present application can better adapt to the regional image features through adaptive brightness partitioning, and achieve more fine enhancement effect; and the enhancement method of the present application realizes more accurate contrast enhancement while better preserving the structural information of the image through adaptive brightness partitioning and variational model, far exceeding other image enhancement methods at the present stage.

[0116] The above-described embodiments only express several embodiments of the present application, which are described in detail and specifically, but should not be understood as limiting the scope of the patent. It should be noted that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the scope of protection of the present application patent should be subject to the appended claims.

Claims

1. A low-light image enhancement method, characterized in that, Includes the following steps: Acquire a low-light image and the brightness value of each pixel in the low-light image, and calculate the average brightness value of all pixels in the low-light image; A brightness threshold is set based on the average brightness value, and the image domain of the low-light image is divided into dark areas and bright areas based on the brightness threshold; where bright areas are areas with brightness values ​​higher than the brightness threshold, and dark areas are areas with brightness values ​​lower than the brightness threshold. A contrast enhancement model is constructed, and the brightness of each pixel in the dark area is enhanced based on the contrast enhancement model to obtain an image with enhanced brightness; the contrast enhancement model is represented as: ； in: This represents the gradient consistency constraint term. Indicates the output image gradient, Let Ω represent the gradient of the adaptive gradient adjustment function, and let Ω represent the image domain. Represents the regularization term. Represents the balance parameters. This represents the adaptive brightness adjustment function; The feature function is represented by the gradient consistency constraint term, which is used to adjust the differential structure of each pixel in the dark area, including edge features and texture features, in order to adjust the contrast of each pixel in the dark area. The regularization term is used to enhance the brightness of each pixel in the dark area. The feature function is used to constrain the pixel value of each pixel in the dark area. The adaptive gradient adjustment function within the contrast enhancement model Represented as: ； The adaptive brightness adjustment function within the contrast enhancement model Represented as: ； in: α Indicates the contrast parameter; β Indicates brightness parameter; Represents the bright areas of the image domain in a low-light image; This represents the dark area in the image domain of a low-light image; The characteristic function Represented as: ； in: S This represents a closed bounded interval.

2. The low-light image enhancement method according to claim 1, characterized in that, When the low-light image is a grayscale image, the image domain of the grayscale image is divided into dark areas and bright areas, including: Obtain the brightness value of each pixel in the grayscale image, and calculate the average brightness value of all pixels in the grayscale image. ; Based on average brightness value Set a brightness threshold for the grayscale image, which is 1.1 times the average brightness value. The image domain of a grayscale image is divided into dark and bright areas based on a brightness threshold. In this context, bright areas within a grayscale image are regions with a brightness threshold higher than the brightness threshold, while dark areas within a grayscale image are regions with a brightness threshold lower than the brightness threshold.

3. The low-light image enhancement method according to claim 1, characterized in that, When the low-light image is a color image, the image domain of the color image is divided into dark areas and bright areas, including: Take each pixel in the color image and decompose each pixel into three-channel components. , and Get the maximum channel luminance value for each pixel. , is represented as: ； Calculate the global average maximum brightness value for all pixels. , is represented as: ； in: x ∈Ω; Based on the global average maximum brightness value Set a brightness threshold for the color image, which is 1.1 times the global average maximum brightness value. The image domain of a color image is divided into dark and bright areas based on a brightness threshold. In this context, bright areas in a color image are regions with a brightness threshold higher than the brightness threshold, while dark areas in a color image are regions with a brightness threshold lower than the brightness threshold.

4. The low-light image enhancement method according to claim 3, characterized in that, Setting the brightness threshold of the color image includes: Maximum channel brightness value Proportional to the three channels represented by hue, saturation, and brightness HSV, the normalized expression is: ； but V With maximum channel brightness value The division is consistent.

5. A low-light image enhancement device, characterized in that, include: The image partitioning module is used to acquire low-light images and the brightness value of each pixel in the low-light images, and to calculate the average brightness value of all pixels in the low-light images. A brightness threshold is set based on the average brightness value, and the image domain of the low-light image is divided into dark areas and bright areas based on the brightness threshold; where bright areas are areas with brightness values ​​higher than the brightness threshold, and dark areas are areas with brightness values ​​lower than the brightness threshold. The image enhancement module is used to construct a contrast enhancement model, and based on the contrast enhancement model, enhance the brightness of each pixel in the dark area to obtain an image with enhanced brightness; the contrast enhancement model is represented as: ； in: This represents the gradient consistency constraint term. Indicates the output image gradient, Let Ω represent the gradient of the adaptive gradient adjustment function, and let Ω represent the image domain. Represents the regularization term. Represents the balance parameters. This represents the adaptive brightness adjustment function; The feature function is represented by the gradient consistency constraint term, which is used to adjust the differential structure of each pixel in the dark area, including edge features and texture features, in order to adjust the contrast of each pixel in the dark area. The regularization term is used to enhance the brightness of each pixel in the dark area. The feature function is used to constrain the pixel value of each pixel in the dark area. The adaptive gradient adjustment function within the contrast enhancement model Represented as: ； The adaptive brightness adjustment function within the contrast enhancement model Represented as: ； in: α Indicates the contrast parameter; β Indicates brightness parameter; Represents the bright areas of the image domain in a low-light image; This represents the dark area in the image domain of a low-light image; The characteristic function Represented as: ； in: S This represents a closed bounded interval.

6. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the steps of the low-light image enhancement method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the steps of a low-light image enhancement method as described in any one of claims 1 to 4.

Images