Low-light image enhancement method and device

Abstract

The invention relates to the field of image enhancement, provides a low-light image enhancement method and device, aims to realize low-light image enhancement with interpretability, low complexity and high enhancement performance, and realizes differential enhancement of different brightness intervals by constructing a bipartite function enhancement model and utilizing correction parameters in the bipartite function enhancement model to regulate and control concavity and convexity of a function curve. Meanwhile, parameters are optimized in combination with a genetic algorithm, the defects that a traditional method is poor in enhancement effect and parameters depend on manual adjustment and optimization are overcome, the problems that a deep learning method is high in complexity and weak in interpretability are solved, core indexes such as detail reservation, color naturalness, real-time performance and scene adaptability of the low-light image are obviously improved, and the method is suitable for popularization and application. The method can be widely applied to the fields of video monitoring, automatic driving, medical imaging, computational photography, industrial visual inspection and the like.

Description

Technical Field

[0001] This invention relates to the field of image enhancement, specifically a method and apparatus for enhancing low-light images. Background Technology

[0002] Low-light images are images obtained under conditions of insufficient light intensity, typically exhibiting low brightness, blurred texture details, and low contrast. Low-light image enhancement, a crucial research direction in computer vision, aims to improve the visual quality and interpretability of images captured under low-light conditions. Existing low-light image enhancement methods are mainly divided into classical methods and deep learning-based methods: Classical methods, such as adaptive gamma correction and Retinex theory-derived methods, while possessing some interpretability, have limitations in dynamic range processing and contrast balance, making them difficult to handle complex low-light scenes; Deep learning-based methods, such as convolutional neural networks and Transformer models, while performing well in some benchmark tests, suffer from "black box" characteristics, high model complexity, high computational cost, and still fall short in balancing noise suppression and detail preservation.

[0003] Furthermore, existing methods generally suffer from problems such as parameter dependence on manual tuning, weak generalization ability, and difficulty in balancing enhancement effect and computational efficiency. Therefore, there is an urgent need for a low-light image enhancement method that combines interpretability, low complexity, and high enhancement performance to meet the real-time and stability requirements of practical applications. Summary of the Invention

[0004] To achieve low-light image enhancement that is interpretable, low-complexity, and high-enhancement performance, this invention provides a low-light image enhancement method and apparatus.

[0005] The technical solution adopted by the present invention to solve the above problems is:

[0006] Low-light image enhancement methods include:

[0007] Step 1: Normalize the pixel values ​​of the image to be processed to obtain image img0;

[0008] Step 2: Convert the image img0 from the RGB color space to the HSV color space, and separate the hue component H, saturation component S, and brightness component V.

[0009] Step 3: After enhancing the luminance component V based on the preset bisection function, normalization is performed simultaneously to obtain... The expression for the binary search function is: ,in Input luminance component pixel values, To correct the parameters, Scaling factor The enhanced luminance component;

[0010] Step 4, based on H, S, V and Color restoration yields image img1;

[0011] Step 5: Perform noise suppression on image img1 to obtain image img2;

[0012] Step 6: Merge images img0, img1, and img2 to obtain img3;

[0013] Step 7: Post-process img3 to obtain the final enhanced image.

[0014] Furthermore, step 4 specifically involves:

[0015] Step 41: Convert the luminance component and inverse value Multiply and normalize to obtain ;

[0016] Step 42, using the formula The recovered luminance components were calculated. At the same time, Normalization is performed, where a and b are color recovery parameters;

[0017] Step 43: Combine H, S and The image is reconstructed into an HSV image and then converted to the RGB color space to obtain the color-restored image img1.

[0018] Furthermore, step 5 directly employs guided filtering for noise suppression or

[0019] Convert img1 to the YCbCr color space, separate the luminance channel Y, chrominance channels Cb and Cr, apply guided filtering to the Y channel, and then combine the filtered Y channel with the Cb and Cr channels to form a YCbCr image, which is then converted to the RGB color space to obtain img2.

[0020] Furthermore, the guided filtering implementation process includes: determining the filter window radius. and regularization parameters Within the RGB color space, calculate the mean of each channel of the guide image and the input image within the filtering window; calculate the variance of the guide image and the covariance of the guide image and the input image; calculate the filtering coefficients m and n based on the covariance and variance, and calculate the mean of m and n within the filtering window; calculate the filtering result based on the mean of m and n and the guide image to obtain the filtered image.

[0021] Further post-processing includes: normalization, gamma correction, scaling the corrected image pixel values ​​to the range of [0, 255], and converting them to uint8 type.

[0022] Furthermore, the function expression used for normalization is: ,in For image data to be normalized, The minimum value of the image. The maximum value of the image, when When the time is 0, the normalization result is 0.

[0023] Furthermore, in step 6, , where c, d, and e are fusion parameters.

[0024] Furthermore, the color restoration parameters and fusion parameters are determined based on genetic algorithm optimization.

[0025] A low-light image enhancement apparatus for implementing a low-light image enhancement method, comprising:

[0026] Image reading module: used to read input images;

[0027] Normalization module: Used to perform image normalization processing;

[0028] Color space conversion module: used to convert an image from the RGB color space to the HSV color space and separate the hue component (H), saturation component (S), and lightness component (V);

[0029] Bisection function enhancement module: Based on a preset bisection function, the luminance component V is enhanced and then normalized to obtain... ;

[0030] Color restoration module: based on H, S, V and Perform color restoration;

[0031] Noise suppression module: Used to perform noise suppression processing on images;

[0032] Image fusion module: used for image fusion;

[0033] Final enhancement module: Post-processes the fused image to obtain the final enhanced image.

[0034] Furthermore, it also includes a parameter optimization module: used to perform the genetic algorithm optimization process.

[0035] The advantages of this invention compared to existing technologies are: by constructing a bisection function enhancement model, and utilizing the correction parameters therein... By adjusting the concavity and convexity of the control function curve, differentiated enhancements are achieved for different brightness ranges. At the same time, the parameters are optimized by combining genetic algorithms. This not only overcomes the shortcomings of traditional methods, such as poor enhancement effect and dependence on manual parameter tuning, but also solves the problems of high complexity and weak interpretability of deep learning methods. Significant improvements have been achieved in key indicators such as detail preservation, color naturalness, real-time performance, and scene adaptability of low-light images. It can be widely used in fields such as video surveillance, autonomous driving, medical imaging, computational photography, and industrial vision inspection. Attached Figure Description

[0036] Figure 1 This is a flowchart of a low-light image enhancement method. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0038] like Figure 1 As shown, the low-light image enhancement method includes:

[0039] Step 1: Normalize the pixel values ​​of the image to be processed to obtain image img0.

[0040] The image to be processed is a low-light image in RGB format, with each pixel value ranging from [0, 255]. Due to different actual shooting scenes, the image may have differences in brightness, contrast, etc. To reduce the adverse effects of such differences, the pixel values ​​are normalized to the range of [0, 1]. Specifically, a function is used... Normalize the low-light image, where For image data to be normalized, The minimum value of the image. The maximum value of the image, when When the time is 0, the normalization result is 0.

[0041] Step 2: Convert the image img0 from the RGB color space to the HSV color space, and separate the hue component H, saturation component S, and brightness component V.

[0042] To better align with human color perception, the image is converted from the RGB color space to the HSV color space. The conversion function is as follows:

[0043] ,

[0044] ,

[0045] ,

[0046] Where r, g, and b represent the red, green, and blue components of the corresponding pixel, respectively, max is the maximum value of the three components, and min is the minimum value of the three components.

[0047] Step 3: After enhancing the luminance component V based on the preset bisection function, normalization is performed simultaneously to obtain... The expression for the binary search function is: ,in Input luminance component pixel values, The calibration parameters can be determined through optimization using a genetic algorithm. This is the scaling factor, with a value range of [0,1]. This is the enhanced luminance component.

[0048] Step 4, based on H, S, V and Color restoration yields image img1.

[0049] Brightness component and inverse value Multiply and normalize to obtain Then, through the formula The recovered luminance components were calculated. At the same time, the same normalization function as in step S1 is used to process... Normalization is performed, where a and b are color recovery parameters, which can be determined through genetic algorithm optimization; finally, H and S are... The image is reconstructed into an HSV image and then converted to the RGB color space to obtain the color-restored image img1.

[0050] Step 5: Perform noise suppression on image img1 to obtain image img2.

[0051] Noise suppression is performed on img1 using guided filtering: First, the radius of the filtering window is determined. and regularization parameters A preferred configuration is , Then, in the RGB color space, the mean of each channel of the guide image and the input image within the filtering window is calculated; then the variance of the guide image and the covariance of the guide image and the input image are calculated; then the filtering coefficients m and n are calculated based on the covariance and variance, and the mean of m and n within the filtering window is calculated; finally, the filtering result is calculated based on the mean of m and n and the guide image to obtain the filtered image img2.

[0052] Noise suppression can also be achieved by an alternative approach: converting img1 to the YCbCr color space, separating the luminance channel Y, chrominance channels Cb and Cr, applying guided filtering to the Y channel, and then combining the filtered Y channel with the Cb and Cr channels to form a YCbCr image, which is then converted to the RGB color space to obtain img2.

[0053] Step 6: Fuse images img0, img1, and img2 to obtain img3.

[0054] based on Image fusion is performed, and img3 is normalized. c, d, and e are fusion parameters that can be determined by optimization using a genetic algorithm.

[0055] Step 7: Post-process img3 to obtain the final enhanced image.

[0056] Gamma correction was performed on img3, with the correction parameters being: The corrected image pixel values ​​are scaled to the range [0, 255] and converted to uint8 type to obtain the final enhanced image. (Parameters...) It can be determined through optimization using a genetic algorithm.

[0057] The genetic algorithm optimization process includes:

[0058] (1) Parameter initialization: Determine a, b, c, d, e, , The range of values ​​for is used to generate an initial parameter population of size N, where N is a preset positive integer;

[0059] In practice, the values ​​of a, b, c, d, and e are set to the range [0.1, 0.9]. The value range is set to [0.2, 2.0]. The value range is set to [0.8, 1.5], and N is set to 10.

[0060] (2) Fitness calculation: For each parameter combination in the population, the enhanced image is obtained through steps 1-7. The PSNR and SSIM of the enhanced image and the ground truth image are calculated based on the formula. Calculate fitness, where , , To enhance the image, This is a true value image;

[0061] (3) Selection operation: The roulette wheel selection method is adopted. The selection probability of each parameter combination is determined according to the fitness, and a new population is generated.

[0062] (4) Crossover operation: A simulated binary crossover operator is used to crossover the parameter combinations in the new population to generate offspring parameter combinations. The crossover process is based on random variables. and preset distribution parameters Calculate the expansion factor Thus, the offspring parameters are obtained;

[0063] In practical implementation, random variables For randomly generated floating-point numbers in the range [0,1], the distribution parameters are... Set to 2, expansion factor The calculation formula is:

[0064] .

[0065] (5) Mutation operation: The offspring parameter combination is mutated using a polynomial mutation operator based on random variables. and preset distribution parameters Calculate the variation expansion factor Update the parameters and ensure they are within their range of values;

[0066] In practical implementation, random variables For randomly generated floating-point numbers in the range [0,1], the distribution parameters are... Set to 25, expansion factor The calculation formula is:

[0067] .

[0068] in The difference between the value generated by the previous crossover operation and the minimum value of the parameter's set range, divided by the size of the parameter's set range, is calculated as follows: The difference between the maximum value of the parameter's range and the value generated by the previous cross operation is divided by the range size set for the parameter.

[0069] (6) Iterative update: Repeat steps (2)-(5) until the preset number of iterations (50 times) is reached, and select the parameter combination with the highest fitness as the optimal parameter.

[0070] Correspondingly, this embodiment also provides a low-light image enhancement apparatus for implementing a low-light image enhancement method, including:

[0071] Image reading module: used to read input images;

[0072] Normalization module: Used to perform image normalization processing;

[0073] Color space conversion module: used to convert an image from the RGB color space to the HSV color space and separate the hue component (H), saturation component (S), and lightness component (V);

[0074] Bisection function enhancement module: Based on a preset bisection function, the luminance component V is enhanced and then normalized to obtain... ;

[0075] Color restoration module: based on H, S, V and Perform color restoration;

[0076] Noise suppression module: Used to perform noise suppression processing on images;

[0077] Image fusion module: used for image fusion;

[0078] Final enhancement module: Post-processes the fused image to obtain the final enhanced image.

[0079] Furthermore, it also includes a parameter optimization module: used to perform the genetic algorithm optimization process. During parameter optimization, the input image also includes the ground truth image.

[0080] Enhancement based on bisection functions overcomes the shortcomings of traditional methods, such as poor enhancement effect and dependence on manual parameters, and solves the problems of high complexity and weak interpretability of deep learning methods. It achieves significant improvements in core indicators such as detail preservation, color naturalness, real-time performance, and scene adaptability of low-light images, and can be widely used in fields such as video surveillance, autonomous driving, medical imaging, computational photography, and industrial vision inspection.

Claims

1. A low-light image enhancement method, characterized in that, include: Step 1: Normalize the pixel values ​​of the image to be processed to obtain image img0; Step 2: Convert the image img0 from the RGB color space to the HSV color space, and separate the hue component H, saturation component S, and brightness component V. Step 3: After enhancing the luminance component V based on the preset bisection function, normalization is performed simultaneously to obtain... The expression for the binary search function is: ,in Input luminance component pixel values, To correct the parameters, Scaling factor The enhanced luminance component; Step 4, based on H, S, V and Color restoration yields image img1; Step 5: Perform noise suppression on image img1 to obtain image img2; Step 6: Merge images img0, img1, and img2 to obtain img3; Step 7: Post-process img3 to obtain the final enhanced image.

2. The low-light image enhancement method according to claim 1, characterized in that, Step 4 is as follows: Step 41: Convert the luminance component and inverse value Multiply and normalize to obtain ; Step 42, using the formula The recovered luminance components were calculated. At the same time, Normalization is performed, where a and b are color recovery parameters; Step 43: Combine H, S and The image is reconstructed into an HSV image and then converted to the RGB color space to obtain the color-restored image img1.

3. The low-light image enhancement method according to claim 1, characterized in that, Step 5 directly uses guided filtering for noise suppression or Convert img1 to the YCbCr color space, separate the luminance channel Y, chrominance channels Cb and Cr, apply guided filtering to the Y channel, and then combine the filtered Y channel with the Cb and Cr channels to form a YCbCr image, which is then converted to the RGB color space to obtain img2.

4. The low-light image enhancement method according to claim 3, characterized in that, The guided filtering process includes: determining the filter window radius. and regularization parameters Within the RGB color space, calculate the mean of each channel of the guide image and the input image within the filtering window; calculate the variance of the guide image and the covariance of the guide image and the input image; calculate the filtering coefficients m and n based on the covariance and variance, and calculate the mean of m and n within the filtering window; calculate the filtering result based on the mean of m and n and the guide image to obtain the filtered image.

5. The low-light image enhancement method according to claim 1, characterized in that, Post-processing includes: normalization, gamma correction, scaling the corrected image pixel values ​​to the range of [0, 255], and converting them to uint8 type.

6. The low-light image enhancement method according to claim 1, characterized in that, The function expression used for normalization is: ,in For image data to be normalized, The minimum value of the image. The maximum value of the image, when When the time is 0, the normalization result is 0.

7. The low-light image enhancement method according to claim 1, characterized in that, In step 6, , where c, d, and e are fusion parameters.

8. The low-light image enhancement method according to claim 7, characterized in that, Color restoration and fusion parameters are determined based on genetic algorithm optimization.

9. A low-light image enhancement apparatus, used to implement the low-light image enhancement method according to any one of claims 1-8, characterized in that, include: Image reading module: used to read input images; Normalization module: Used to perform image normalization processing; Color space conversion module: used to convert an image from the RGB color space to the HSV color space and separate the hue component (H), saturation component (S), and lightness component (V); Bisection function enhancement module: Based on a preset bisection function, the luminance component V is enhanced and then normalized to obtain... ; Color restoration module: based on H, S, V and Perform color restoration; Noise suppression module: Used to perform noise suppression processing on images; Image fusion module: used for image fusion; Final enhancement module: Post-processes the fused image to obtain the final enhanced image.

10. The low-light image enhancement apparatus according to claim 9, characterized in that, It also includes a parameter optimization module: used to perform the genetic algorithm optimization process.

Images