Image adaptive enhancement method and system based on linear dynamic weight fusion

Abstract

The invention discloses an image adaptive enhancement method and system based on linear dynamic weight fusion, and belongs to the technical field of image enhancement. The brightness value of each pixel in an input image is calculated by using human visual characteristics and RGB channel information, and then the global brightness of the input image is calculated based on the brightness value of each pixel in the input image; comparing the global brightness of the input image with a set brightness threshold value, and calculating the local brightness standard deviation of each pixel in the input image to obtain the local brightness standard deviation; according to the method, the functions of judging image illumination features and dynamically matching and enhancing by adopting an overall and local brightness two-dimensional perception mechanism are realized, and the enhancement intensity can be adaptively adjusted according to the regional noise level by adopting a noise perception mechanism based on local variance, so that the balance of detail enhancement and noise suppression is ensured; and the enhancement precision and robustness in a complex low-illumination scene are also significantly improved.

Description

Technical Field

[0001] This invention relates to the field of image enhancement technology, and specifically to an image adaptive enhancement method and system based on linear dynamic weight fusion. Background Technology

[0002] Image enhancement technology has become a key tool in modern military, security, and civilian fields. Its core significance lies in expanding the boundaries of human perception in dark environments, enhancing security, efficiency, and exploration capabilities. As technology advances, its application scenarios will continue to expand, but a balance must be struck between technological innovation and ethical and privacy concerns.

[0003] Currently, existing nighttime image enhancement technologies generally rely on manually preset enhancement intensity, filter kernel size, or brightness threshold parameters. This leads to overexposure or underexposure in complex scenes such as bright light sources, shadow occlusion, and low-light detail areas. Furthermore, these technologies suffer from limitations in adaptability and computational efficiency in complex low-light scenarios. A single global enhancement strategy struggles to balance the processing needs of different brightness regions, especially in areas where bright light sources and low-light details overlap, where distortion and failure are common. The lack of a noise perception mechanism means that existing nighttime image enhancement technologies tend to amplify original noise when enhancing low-light areas, resulting in a significant reduction in the image signal-to-noise ratio. This not only fails to accurately match the enhancement needs of different brightness regions but also faces performance bottlenecks such as high computational complexity and poor real-time performance when processing batch nighttime images. Therefore, it is necessary to design an image adaptive enhancement method and system based on linear dynamic weight fusion. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and to better and more effectively address the limitations of current nighttime image enhancement techniques. These techniques generally rely on manually preset enhancement intensity, filter kernel size, or brightness threshold parameters, leading to overexposure or underexposure in complex scenes such as high-brightness light sources, shadow occlusion, and low-light detail areas. Furthermore, they suffer from adaptability and computational efficiency bottlenecks in complex low-light scenes. A single global enhancement strategy struggles to balance the processing needs of different brightness areas, especially in areas where high-brightness light sources and low-light detail overlap, where distortion and failure are common. The lack of a noise perception mechanism further exacerbates the limitations of existing nighttime image enhancement techniques, which are often prone to errors when enhancing low-light areas. Original noise is easily amplified, resulting in a significant decrease in the signal-to-noise ratio of the image. This not only fails to accurately match the enhancement needs of different brightness areas, but also faces performance bottlenecks such as high computational complexity and poor real-time performance when processing batch night images. To address this, we propose an image adaptive enhancement method and system based on linear dynamic weight fusion. This method achieves the function of judging image illumination characteristics and dynamically matching enhancement by using a dual-dimensional perception mechanism of overall and local brightness. Furthermore, by adopting a noise perception mechanism based on local variance, it can adaptively adjust the enhancement intensity according to the regional noise level. This not only ensures a balance between detail enhancement and noise suppression, but also significantly improves the enhancement accuracy and robustness in complex low-light scenes.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: An image adaptive enhancement method based on linear dynamic weight fusion includes the following steps: Step A: Calculate the brightness value of each pixel in the input image using the characteristics of human vision and RGB channel information, and then calculate the global brightness of the input image based on the brightness value of each pixel in the input image. Step B: Compare the global brightness of the input image with the set brightness threshold. If the global brightness of the input image is lower than the set brightness threshold, continue image enhancement. If the global brightness of the input image is higher than the set brightness threshold, output the input image directly. Step C: Calculate the local brightness standard deviation of each pixel in the input image and obtain the local brightness standard deviation, and then calculate the global brightness variance based on the local brightness standard deviation. Step D: Calculate the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, and then calculate the weight values ​​of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation to obtain the spatial domain Gaussian kernel matrix. Step E: Perform a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix. Then, perform element-wise multiplication in the frequency domain on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result. Step F: Perform a two-dimensional inverse fast Fourier transform on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component. Then, use the local brightness standard deviation to calculate the global brightness standard deviation threshold. Step G: Compare the local brightness standard deviation of each pixel with the global brightness standard deviation threshold to obtain the standard deviation comparison result, and then process the first image illumination component of each pixel according to the standard deviation comparison result to obtain the second image illumination component. Step H: Calculate the color channel reflectance component using the illumination components of the original and second images of the input image, then normalize the color channel reflectance component and obtain the detail enhancement term. Step 1: Based on the brightness values ​​of each pixel in the input image, perform image enhancement using detail enhancement to obtain the enhanced image, and then output the enhanced image to complete the image enhancement operation.

[0006] The aforementioned image adaptive enhancement method based on linear dynamic weight fusion, step A, calculates the brightness value of each pixel in the input image using human visual characteristics and RGB channel information, and then calculates the global brightness of the input image based on the brightness values ​​of each pixel in the input image. The specific steps are as follows. Step A1: Calculate the brightness value of each pixel in the input image using the characteristics of human visual perception and RGB channel information, as shown in formula (1). (1) in, For the pixels in the input image The corresponding brightness value, the pixel For any pixel in the input image, and pixels The number of rows and columns, , and These are the sensitivity coefficients of the human eye to red, green, and blue, respectively. , and pixels Normalized pixel values ​​in the red, green, and blue channels; Step A2: Calculate the global brightness of the input image based on the brightness values ​​of each pixel in the input image, as shown in formula (2). (2) in, To input the global brightness of the image, and These represent the total number of rows and columns of pixels in the input image, respectively.

[0007] The aforementioned image adaptive enhancement method based on linear dynamic weight fusion, in step C, calculates the local brightness standard deviation of each pixel in the input image and obtains the local brightness standard deviation, and then calculates the global brightness variance based on the local brightness standard deviation. The specific steps are as follows. Step C1: Calculate and obtain the local brightness standard deviation of each pixel in the input image, as shown in formula (3). (3) in, For the pixels in the input image The local brightness standard deviation To input the pixels in the image Pixels in a 3x3 grid centered The brightness of the pixel To input the pixels in the image Any pixel in a 3×3 grid centered on the center and pixels The number of rows and columns, For all pixels The mean, For pixels The total number; Step C2: Calculate the global luminance variance based on the local luminance standard deviation, as shown in formula (4). (4) in, This represents the global brightness variance.

[0008] The aforementioned image adaptive enhancement method based on linear dynamic weight fusion, in step D, calculates the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, and then calculates the weight values ​​of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation to obtain the spatial domain Gaussian kernel matrix. The specific steps are as follows. Step D1: Calculate the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, as shown in formula (5). (5) in, The standard deviation of the Gaussian kernel. and These are the maximum and minimum values ​​of the Gaussian kernel standard deviation, respectively. To adjust the coefficient, This represents the maximum global brightness variance. Step D2: Calculate the weight values ​​of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation and obtain the spatial domain Gaussian kernel matrix, as shown in formula (6). (6) in, For the Gaussian nucleus relative to the nucleus center pixels Weight value, the pixel For the Gaussian nucleus relative to the nucleus center The coordinates of any pixel, The coordinates are the center coordinates of the Gaussian kernel.

[0009] The aforementioned image adaptive enhancement method based on linear dynamic weight fusion, in step E, involves performing a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix. Then, the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix are multiplied element-wise in the frequency domain to obtain the frequency domain processing result. The specific steps are as follows. Step E1 involves performing a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix, as shown in formula (7). ; (7) in, This is the transformed color channel matrix. For frequency domain coordinates, The input image color channel matrix, the input image color channel matrix It includes a red channel matrix, a green channel matrix, and a blue channel matrix. The transformed spatial domain Gaussian kernel matrix, For the spatial domain Gaussian kernel matrix, The imaginary unit; Step E2 involves performing element-wise multiplication in the frequency domain on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result, as shown in formula (8). (8) in, This is the result of frequency domain processing.

[0010] The aforementioned image adaptive enhancement method based on linear dynamic weight fusion, in step F, involves performing a two-dimensional inverse fast Fourier transform on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component. Then, the global brightness standard deviation threshold is calculated using the local brightness standard deviation. The specific steps are as follows. Step F1 involves performing a two-dimensional inverse fast Fourier transform on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component, as shown in formula (9). (9) in, This is the first image illumination component; Step F1: Calculate the global brightness standard deviation threshold using the local brightness standard deviation, as shown in formula (10). (10) in, This is the global brightness standard deviation threshold.

[0011] In the aforementioned image adaptive enhancement method based on linear dynamic weight fusion, step G involves comparing the local brightness standard deviation of each pixel with the global brightness standard deviation threshold to obtain the standard deviation comparison result. Then, based on the standard deviation comparison result, the first image illumination component of each pixel is processed to obtain the second image illumination component. Specifically, if the local brightness standard deviation of a pixel is less than the global brightness standard deviation threshold, then the second image illumination component of that pixel is equal to the first image illumination component. If the local brightness standard deviation of a pixel is greater than the global brightness standard deviation threshold, then the second image illumination component of that pixel is half of the first image illumination component.

[0012] The aforementioned image adaptive enhancement method based on linear dynamic weight fusion, in step H, calculates the color channel reflectance component using the illumination components of the original and second input images, then normalizes the color channel reflectance component to obtain the detail enhancement term. The specific steps are as follows: Step H1: Calculate the color channel reflectance component using the illumination components of the original input image and the second image, as shown in formula (11). (11) in, For the color channel reflection component, To reduce the ellipsis factor, if the illumination component of the second image is equal to the illumination component of the first image, then If the second image illumination component is 1, and the second image illumination component is half the first image illumination component, then It is 0.5; Step H2 involves normalizing the color channel reflection components and obtaining the detail enhancement term, as shown in formula (12). (12) in, This is for detail enhancements.

[0013] The aforementioned image adaptive enhancement method based on linear dynamic weight fusion, in step I, uses a detail enhancement term to enhance the image based on the brightness value of each pixel in the input image and obtains the enhanced image. Then, the enhanced image is output to complete the image enhancement operation, as shown in formula (13). (13) in, To enhance the post-image.

[0014] The image adaptive enhancement system based on linear dynamic weight fusion includes a brightness calculation module, a brightness contrast module, a variance calculation module, a Gaussian kernel matrix acquisition module, a frequency domain processing module, an illumination component calculation module, an illumination component processing module, a detail enhancement item acquisition module, and an image enhancement module. The brightness calculation module is used to calculate the brightness value of each pixel in the input image using the characteristics of human visual vision and RGB channel information, and then calculate the global brightness of the input image based on the brightness value of each pixel in the input image. The brightness comparison module is used to compare the global brightness of the input image with a set brightness threshold. If the global brightness of the input image is lower than the set brightness threshold, image enhancement will continue. If the global brightness of the input image is higher than the set brightness threshold, the input image will be output directly. The variance calculation module is used to calculate and obtain the local brightness standard deviation of each pixel in the input image, and then calculate the global brightness variance based on the local brightness standard deviation. The Gaussian kernel matrix acquisition module is used to calculate the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, and then calculate the weight value of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation to obtain the spatial domain Gaussian kernel matrix. The frequency domain processing module is used to perform a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix, and then perform element-wise frequency domain multiplication on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result. The illumination component calculation module is used to perform a two-dimensional inverse fast Fourier transform on the frequency domain processing results to map them back to the spatial domain and obtain the first image illumination component, and then use the local brightness standard deviation to calculate the global brightness standard deviation threshold. The illumination component processing module is used to compare the local brightness standard deviation of each pixel with the global brightness standard deviation threshold and obtain the standard deviation comparison result. Then, based on the standard deviation comparison result, the first image illumination component of each pixel is processed to obtain the second image illumination component. The detail enhancement module is used to calculate the color channel reflection component using the illumination components of the original image and the second image of the input image, and then normalize the color channel reflection component to obtain the detail enhancement item. The image enhancement module is used to enhance the image based on the brightness value of each pixel in the input image using detail enhancement terms, and then output the enhanced image to complete the image enhancement operation.

[0015] The beneficial effects of this invention are as follows: The image adaptive enhancement method and system based on linear dynamic weight fusion of this invention first calculates the brightness value of each pixel in the input image using human visual characteristics and RGB channel information. Then, it calculates the global brightness of the input image based on the brightness values ​​of each pixel. Next, it compares the global brightness of the input image with a set brightness threshold. If the global brightness of the input image is lower than the set brightness threshold, image enhancement continues; if the global brightness of the input image is higher than the set brightness threshold, the input image is directly output. Subsequently, the local brightness standard deviation of each pixel in the input image is calculated and obtained. Then, based on the local brightness standard deviation... The standard deviation is used to calculate the global luminance variance. Then, the standard deviation of the Gaussian kernel is calculated using the local luminance standard deviation and the global luminance variance. Based on the Gaussian kernel standard deviation, the weight values ​​of each pixel in the input image within the Gaussian kernel are calculated to obtain the spatial domain Gaussian kernel matrix. Next, a two-dimensional fast Fourier transform is performed on the input image color channel matrix and the spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix. Then, element-wise multiplication in the frequency domain is performed on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result. Finally, a two-dimensional inverse fast Fourier transform is performed on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component. Then, the local luminance standard deviation is used to calculate the weight values ​​of each pixel in the Gaussian kernel matrix and obtain the spatial domain Gaussian kernel matrix. The local luminance standard deviation is used to calculate the global luminance standard deviation threshold. Then, the local luminance standard deviation of each pixel is compared with the global luminance standard deviation threshold to obtain the standard deviation comparison result. Based on the standard deviation comparison result, the first image illumination component of each pixel is processed to obtain the second image illumination component. Finally, the color channel reflectance component is calculated using the original input image and the second image illumination components. The color channel reflectance component is then normalized to obtain a detail enhancement term. Next, based on the luminance values ​​of each pixel in the input image, the detail enhancement term is applied to perform image enhancement to obtain the enhanced image. Finally, the enhanced image is output to complete the image enhancement operation; this effectively achieves the desired image enhancement. The adaptive enhancement method and system have the function of judging the image illumination features and dynamically matching enhancement by adopting a dual-dimensional perception mechanism of overall and local brightness. Furthermore, by adopting a noise perception mechanism based on local variance, it can adaptively adjust the enhancement intensity according to the regional noise level, ensuring a balance between detail enhancement and noise suppression. Moreover, by using a linear dynamic weight fusion method based on HSL color space and combining pixel brightness, saturation, and highlight features to dynamically allocate fusion weights, it can ensure the authenticity of highlight areas. This not only breaks through the technical limitations of traditional methods such as strong parameter dependence, insufficient noise suppression, highlight distortion, and poor real-time performance, but also significantly improves the enhancement accuracy and robustness in complex low-light scenes. Attached Figure Description

[0016] Figure 1 This is an overall flowchart of the image adaptive enhancement method based on linear dynamic weight fusion of the present invention; Figure 2This is a schematic diagram of the image enhancement principle of the image adaptive enhancement system based on linear dynamic weight fusion of the present invention. Figure 3 These are comparison images of image enhancement results with different Gaussian kernel standard deviations in embodiments of the present invention; Figure 4 This is a comparison of different methods used to enhance low-light images in the first image of an embodiment of the present invention; Figure 5 This is a comparison of image enhancement methods used to enhance low-light images in the second image of an embodiment of the present invention; Figure 6 The third image in this embodiment of the invention is a comparison of image enhancement methods used to enhance low-light images. Figure 7 The fourth image in this embodiment of the invention is a comparison of image enhancement methods used to enhance low-light images. Figure 8 The fifth image in this embodiment of the invention is a comparison of image enhancement methods used to enhance low-light images. Figure 9 The sixth image in this embodiment of the invention is a comparison of image enhancement methods used to enhance low-light images. Detailed Implementation

[0017] The present invention will now be further described with reference to the accompanying drawings.

[0018] like Figure 1 As shown, the image adaptive enhancement method based on linear dynamic weight fusion of the present invention includes the following steps: Step A involves calculating the brightness value of each pixel in the input image using the characteristics of human vision and RGB channel information, and then calculating the global brightness of the input image based on the brightness values ​​of each pixel. The specific steps are as follows: Step A1: Calculate the brightness value of each pixel in the input image using the characteristics of human visual perception and RGB channel information, as shown in formula (1). (1) in, For the pixels in the input image The corresponding brightness value, the pixel For any pixel in the input image, and pixels The number of rows and columns, , and These are the sensitivity coefficients of the human eye to red, green, and blue, respectively. , and pixels Normalized pixel values ​​in the red, green, and blue channels; Step A2: Calculate the global brightness of the input image based on the brightness values ​​of each pixel in the input image, as shown in formula (2). (2) in, To input the global brightness of the image, and These represent the total number of rows and columns of pixels in the input image, respectively.

[0019] Step B compares the global brightness of the input image with the set brightness threshold. If the global brightness of the input image is lower than the set brightness threshold, image enhancement continues. If the global brightness of the input image is higher than the set brightness threshold, the input image is output directly.

[0020] Step C involves calculating the local brightness standard deviation of each pixel in the input image and then calculating the global brightness variance based on the local brightness standard deviation. The specific steps are as follows. Step C1: Calculate and obtain the local brightness standard deviation of each pixel in the input image, as shown in formula (3). (3) in, For the pixels in the input image The local brightness standard deviation To input the pixels in the image Pixels in a 3x3 grid centered The brightness of the pixel To input the pixels in the image Any pixel in a 3×3 grid centered on the center and pixels The number of rows and columns, For all pixels The mean, For pixels The total number; Step C2: Calculate the global luminance variance based on the local luminance standard deviation, as shown in formula (4). (4) in, This represents the global brightness variance.

[0021] Step D involves calculating the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, then calculating the weight values ​​of each pixel in the Gaussian kernel based on the Gaussian kernel standard deviation to obtain the spatial domain Gaussian kernel matrix. The specific steps are as follows. Step D1: Calculate the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, as shown in formula (5). (5) in, The standard deviation of the Gaussian kernel. and These are the maximum and minimum values ​​of the Gaussian kernel standard deviation, respectively. To adjust the coefficient, This represents the maximum global brightness variance. Step D2: Calculate the weight values ​​of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation and obtain the spatial domain Gaussian kernel matrix, as shown in formula (6). (6) in, For the Gaussian nucleus relative to the nucleus center pixels Weight value, the pixel For the Gaussian nucleus relative to the nucleus center The coordinates of any pixel, The coordinates are the center coordinates of the Gaussian kernel.

[0022] Step E involves performing a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix. Then, the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix are multiplied element-wise in the frequency domain to obtain the frequency domain processing result. The specific steps are as follows. Step E1 involves performing a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix, as shown in formula (7). ; (7) in, This is the transformed color channel matrix. For frequency domain coordinates, The input image color channel matrix, the input image color channel matrix It includes a red channel matrix, a green channel matrix, and a blue channel matrix. The transformed spatial domain Gaussian kernel matrix, For the spatial domain Gaussian kernel matrix, The imaginary unit; Step E2 involves performing element-wise multiplication in the frequency domain on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result, as shown in formula (8). (8) in, This is the result of frequency domain processing.

[0023] Step F involves performing a two-dimensional inverse fast Fourier transform on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component. Then, the global brightness standard deviation threshold is calculated using the local brightness standard deviation. The specific steps are as follows. Step F1 involves performing a two-dimensional inverse fast Fourier transform on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component, as shown in formula (9). (9) in, This is the first image illumination component; Step F1: Calculate the global brightness standard deviation threshold using the local brightness standard deviation, as shown in formula (10). (10) in, This is the global brightness standard deviation threshold.

[0024] Step G involves comparing the local brightness standard deviation of each pixel with the global brightness standard deviation threshold to obtain the standard deviation comparison result. Then, based on the standard deviation comparison result, the first image illumination component of each pixel is processed to obtain the second image illumination component. Specifically, if the local brightness standard deviation of a pixel is less than the global brightness standard deviation threshold, the second image illumination component of that pixel is equal to the first image illumination component. If the local brightness standard deviation of a pixel is greater than the global brightness standard deviation threshold, the second image illumination component of that pixel is half of the first image illumination component.

[0025] Step H involves calculating the color channel reflectance component using the illumination components of the original and second input images, then normalizing the color channel reflectance component and obtaining the detail enhancement term. The specific steps are as follows. Step H1: Calculate the color channel reflectance component using the illumination components of the original input image and the second image, as shown in formula (11). (11) in, For the color channel reflection component, To reduce the ellipsis factor, if the illumination component of the second image is equal to the illumination component of the first image, then If the second image illumination component is 1, and the second image illumination component is half the first image illumination component, then It is 0.5; Step H2 involves normalizing the color channel reflection components and obtaining the detail enhancement term, as shown in formula (12). (12) in, This is for detail enhancements.

[0026] Step I: Based on the brightness values ​​of each pixel in the input image, a detail enhancement term is used to enhance the image and obtain the enhanced image. Then, the enhanced image is output to complete the image enhancement operation, as shown in formula (13). (13) in, To enhance the post-image.

[0027] The image adaptive enhancement system based on linear dynamic weight fusion includes a brightness calculation module, a brightness contrast module, a variance calculation module, a Gaussian kernel matrix acquisition module, a frequency domain processing module, an illumination component calculation module, an illumination component processing module, a detail enhancement item acquisition module, and an image enhancement module. The brightness calculation module is used to calculate the brightness value of each pixel in the input image using the characteristics of human visual vision and RGB channel information, and then calculate the global brightness of the input image based on the brightness value of each pixel in the input image. The brightness comparison module is used to compare the global brightness of the input image with a set brightness threshold. If the global brightness of the input image is lower than the set brightness threshold, image enhancement will continue. If the global brightness of the input image is higher than the set brightness threshold, the input image will be output directly. The variance calculation module is used to calculate and obtain the local brightness standard deviation of each pixel in the input image, and then calculate the global brightness variance based on the local brightness standard deviation. The Gaussian kernel matrix acquisition module is used to calculate the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, and then calculate the weight value of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation to obtain the spatial domain Gaussian kernel matrix. The frequency domain processing module is used to perform a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix, and then perform element-wise frequency domain multiplication on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result. The illumination component calculation module is used to perform a two-dimensional inverse fast Fourier transform on the frequency domain processing results to map them back to the spatial domain and obtain the first image illumination component, and then use the local brightness standard deviation to calculate the global brightness standard deviation threshold. The illumination component processing module is used to compare the local brightness standard deviation of each pixel with the global brightness standard deviation threshold and obtain the standard deviation comparison result. Then, based on the standard deviation comparison result, the first image illumination component of each pixel is processed to obtain the second image illumination component. The detail enhancement module is used to calculate the color channel reflection component using the illumination components of the original image and the second image of the input image, and then normalize the color channel reflection component to obtain the detail enhancement item. The image enhancement module is used to enhance the image based on the brightness value of each pixel in the input image using detail enhancement terms, and then output the enhanced image to complete the image enhancement operation.

[0028] To illustrate the effects of the present invention, a specific embodiment of enhancing the effect using the method of the present invention is described below; (1) Adaptive Gaussian kernel standard deviation It is a core parameter in image enhancement tasks, balancing noise removal, detail preservation, and brightness equalization. When the Gaussian kernel standard deviation... If the value is too small, the smoothing effect of Gaussian filtering is insufficient, leading to excessive amplification of noise in dark areas, causing overexposure and distortion of pixels in bright areas, and compromising the dynamic range of the image. Figure 3 As shown in Figure a; if the Gaussian kernel standard deviation If the value is too large, the filtering process will excessively blur the image texture, causing a severe loss of detail in dark areas. Even if the overall brightness of the dark areas is improved, the loss of detail will result in a flattened effect. Figure 3 As shown in b; in contrast, the Gaussian kernel standard deviation in the method proposed in this embodiment of the present invention... It can achieve precise adaptation to the light distribution in different areas, such as Figure 3 As shown in c, this achieves balanced optimization of global brightness while effectively suppressing noise in dark areas and fully preserving image details.

[0029] (2) To verify the effectiveness of the enhancement method proposed in this invention in low-light image enhancement tasks, this embodiment selects six representative low-light images for enhancement experiments, such as... Figures 4-9 As shown. Among them, Figure 4 and Figure 5 For low-light scenes with light source interference; Figure 6 , Figure 7 , Figure 8 and Figure 9 Low-light images corresponding to different brightness ranges, and Figure 6 , Figure 7 , Figure 8 and Figure 9The brightness values ​​cover the ranges of 0.2-0.4, 0.1-0.2, 0.05-0.1, and 0.01-0.05, respectively.

[0030] This embodiment compares the enhancement results of the proposed method with those of the following algorithms: SSR (Single Scale Retinex), MSR (Multi-Scale Retinex), MSRCR (Multi-Scale Retinex with Color Restoration), HE (Histogram Equalization), SCI (Sample-based Contrastive Image Enhancement), and Retinex-DIP (Retinex-based Deep Image Prior). Figures 4-9 In the diagram, ah represents the original image, the enhanced image of the SSR algorithm, the enhanced image of the MSR algorithm, the enhanced image of the MSRCR algorithm, the enhanced image of the HE algorithm, the enhanced image of the SCI deep learning method, the enhanced image of the Retinex-DIP deep learning method, and the enhanced image of the method proposed in this invention, respectively.

[0031] In the experimental results of this embodiment, the SSR algorithm and the MSR algorithm have a certain enhancement capability for low-brightness images, but in scenes with relatively high brightness, such as... Figure 6 and Figure 7Over-enhancement can easily lead to image whitening; the MSRCR algorithm can achieve full-scene enhancement, but its excessive focus on color restoration results in noticeable color casts in the enhanced image; the HE algorithm performs well in samples with relatively balanced brightness, but performs poorly in complex areas with significant brightness differences. The SCI deep learning method has a relatively balanced enhancement effect, and it improves in terms of over-enhancement compared to the SSR and MSR algorithms, but it still exhibits slight over-enhancement in medium-brightness scenes and leaves shadows in low-brightness areas. The Retinex-DIP deep learning method has a generally better enhancement performance, but it suffers from color cast issues due to high saturation and overexposure in extremely low-brightness scenes. In contrast, the image enhancement method proposed in this invention can be adapted to low-illuminance image enhancement across the entire brightness range. Its enhancement results do not have overexposure or color cast issues, and the color distribution is uniform and natural, conforming to the characteristics of human visual perception.

[0032] (3) In order to further accurately evaluate the advantages and disadvantages of the proposed method compared with other enhancement methods, this embodiment calculates the image quality scores of Natural Image Quality Evaluator (NIQE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM) based on the enhanced image and creates corresponding score data tables as shown in Table 1, Table 2, and Table 3. The lower the Natural Image Quality Evaluator (NIQE) score, the better the naturalness of the image; the higher the Peak Signal-to-Noise Ratio (PSNR) score, the better the image quality; and the closer the SSIM score is to 1, the better the visual quality of the image.

[0033] Table 1. Comparison of NIQE scores for natural image quality by method in 6 images;

[0034] This embodiment analyzes the Natural Image Quality (NIQE) scores. The average NIQE values ​​of the SSR and MSR algorithms are very close, indicating essentially equal performance. Figure 6 , Figure 7 and Figure 8 Both algorithms performed relatively well, but visualization results show that both exhibited over-enhancement in these images, leading to excessive smoothing and consequently higher Natural Image Quality (NIQE) scores. The MSRCR algorithm displayed the highest NIQE value across all six images, indicating that its enhanced images had the worst naturalness and were prone to color distortion and noise amplification, clearly reflecting its poor image quality. The HE algorithm's overall performance fell between the MSR algorithm and the Retinex-Dip deep learning method. Figure 6The SCI deep learning method performs best in the first image, but only moderately in the remaining images. Its average natural image quality (NIQE) is superior to traditional methods and the Retinex-Dip deep learning method, and only slightly better than the method proposed in this invention. Figure 4 and Figure 5 The method performs relatively well, achieving the best naturalness after enhancement for low-light images with light sources. However, as seen in the images, this method's pursuit of naturalness can lead to insufficient enhancement and over-smoothing in some areas. The Retinex-Dip deep learning method outperforms the SSR, MSR, and HE algorithms in average natural image quality (NIQE), but is weaker than the SCI deep learning method and the proposed method, indicating its limited scene adaptability. In some low-light scenes, it is prone to problems such as high saturation and overexposure, resulting in a decrease in naturalness. The proposed method maintains a low NIQE value in all six images without significant performance fluctuations. Compared to the second-best performing SCI deep learning method, the proposed method... Figure 4 , Figure 7 , Figure 8 and Figure 9 The lower NIQE value of the natural image quality in the scene demonstrates the stronger full brightness range adaptation capability of the method proposed in this invention.

[0035] Table 2. Comparison of Peak Signal-to-Noise Ratio (PSNR) Scores for Each Method in 6 Images;

[0036] Peak Signal-to-Noise Ratio (PSNR), as a quality evaluation metric for full-reference images, measures the degree of distortion between the enhanced and reference images, comprehensively considering pixel grayscale deviation, structural consistency, and detail fidelity. Table 2 shows that the PSNR score of the method proposed in this invention is [missing information - likely related to image quality]. Figure 5 It ranked first among all test samples outside of the standard, which fully demonstrates that its enhancement results have lower distortion and better image quality.

[0037] Table 3. Comparison of SSIM scores for each method in 6 images;

[0038] As shown in Table 3, the method proposed in this invention exhibits the highest score in all samples, indicating that the image enhanced by the method proposed in this invention has the best image quality.

[0039] (4) Computation time is another core indicator for evaluating the efficiency of image enhancement methods. In this embodiment, time consumption tests were conducted based on the above test images. After repeated experiments, the average value was taken to obtain the comparison results of the computation efficiency of each method, as shown in Table 4.

[0040] Table 4. Comparison of computational efficiency of various methods in 6 images;

[0041] As shown in Table 4, compared to the SSR algorithm, the method proposed in this invention achieves a better enhancement effect with a small time cost, realizing a balance between efficiency and effect. Compared to the MSR algorithm, the method proposed in this invention has a comparable time consumption and better stability. The HE algorithm has the fastest operation speed due to its extremely simple statistical-mapping logic, but this speed comes at the cost of enhancement effect, and has defects such as overexposure, noise amplification, and color cast, making it difficult to adapt to the diverse illumination distribution of low-light images in aerial survey scenarios. Compared to supervised deep learning methods that rely on large-scale sample training, the method proposed in this invention has a slightly higher inference time, but saves the high cost of sample collection and model training in the early stage. Compared with the Retinex-Dip deep learning method, the method proposed in this invention reduces the time consumption by more than 90%. The method proposed in this invention has a reasonable computation time while ensuring enhancement performance, and has significant potential for engineering application.

[0042] In summary, the image adaptive enhancement method and system based on linear dynamic weight fusion of the present invention first calculates the brightness value of each pixel in the input image using human visual characteristics and RGB channel information. Then, it calculates the global brightness of the input image based on the brightness values ​​of each pixel. Next, it compares the global brightness of the input image with a set brightness threshold. If the global brightness of the input image is lower than the set brightness threshold, image enhancement continues; if the global brightness of the input image is higher than the set brightness threshold, the input image is directly output. Subsequently, it calculates the local brightness standard deviation of each pixel in the input image and obtains the local brightness standard deviation. Then, it calculates... The global brightness variance is calculated, and then the Gaussian kernel standard deviation is calculated using the local brightness standard deviation and the global brightness variance. Based on the Gaussian kernel standard deviation, the weight values ​​of each pixel in the input image within the Gaussian kernel are calculated to obtain the spatial domain Gaussian kernel matrix. Next, a two-dimensional fast Fourier transform is performed on the input image color channel matrix and the spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix. Then, element-wise multiplication in the frequency domain is performed on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result. Finally, a two-dimensional inverse fast Fourier transform is performed on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component. Then, the local brightness variance is used to calculate the weight values ​​of each pixel in the Gaussian kernel matrix, resulting in the spatial domain Gaussian kernel matrix. The standard deviation of the global luminance is calculated, and then the local luminance standard deviation of each pixel is compared with the global luminance standard deviation threshold to obtain the standard deviation comparison result. Based on the standard deviation comparison result, the first image illumination component of each pixel is processed to obtain the second image illumination component. Finally, the color channel reflectance component is calculated using the original input image and the second image illumination components. The color channel reflectance component is then normalized to obtain a detail enhancement term. Next, based on the luminance values ​​of each pixel in the input image, the detail enhancement term is applied to perform image enhancement to obtain the enhanced image. Finally, the enhanced image is output to complete the image enhancement operation; this effectively realizes the image enhancement... The adaptive enhancement method and system have the function of judging image illumination features and dynamically matching enhancement by adopting a dual-dimensional perception mechanism of overall and local brightness. Furthermore, by adopting a noise perception mechanism based on local variance, the enhancement intensity can be adaptively adjusted according to the regional noise level, ensuring a balance between detail enhancement and noise suppression. Moreover, by using a linear dynamic weight fusion method based on HSL color space and combining pixel brightness, saturation, and high brightness features to dynamically allocate fusion weights, the authenticity of high brightness areas can be guaranteed. This not only breaks through the technical limitations of traditional methods such as strong parameter dependence, insufficient noise suppression, high brightness distortion, and poor real-time performance, but also significantly improves the enhancement accuracy and robustness in complex low-light scenes.

Claims

1. An image adaptive enhancement method based on linear dynamic weight fusion, characterized in that: Includes the following steps, Step A: Calculate the brightness value of each pixel in the input image using the characteristics of human vision and RGB channel information, and then calculate the global brightness of the input image based on the brightness value of each pixel in the input image. Step B: Compare the global brightness of the input image with the set brightness threshold. If the global brightness of the input image is lower than the set brightness threshold, continue image enhancement. If the global brightness of the input image is higher than the set brightness threshold, output the input image directly. Step C: Calculate the local brightness standard deviation of each pixel in the input image and obtain the local brightness standard deviation, and then calculate the global brightness variance based on the local brightness standard deviation. Step D: Calculate the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, and then calculate the weight values ​​of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation to obtain the spatial domain Gaussian kernel matrix. Step E: Perform a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix. Then, perform element-wise multiplication in the frequency domain on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result. Step F: Perform a two-dimensional inverse fast Fourier transform on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component. Then, use the local brightness standard deviation to calculate the global brightness standard deviation threshold. Step G: Compare the local brightness standard deviation of each pixel with the global brightness standard deviation threshold to obtain the standard deviation comparison result, and then process the first image illumination component of each pixel according to the standard deviation comparison result to obtain the second image illumination component. Step H: Calculate the color channel reflectance component using the illumination components of the original and second images of the input image, then normalize the color channel reflectance component and obtain the detail enhancement term. Step 1: Based on the brightness values ​​of each pixel in the input image, perform image enhancement using detail enhancement to obtain the enhanced image, and then output the enhanced image to complete the image enhancement operation.

2. The image adaptive enhancement method based on linear dynamic weight fusion according to claim 1, characterized in that: Step A involves calculating the brightness value of each pixel in the input image using the characteristics of human vision and RGB channel information, and then calculating the global brightness of the input image based on the brightness values ​​of each pixel. The specific steps are as follows: Step A1: Calculate the brightness value of each pixel in the input image using the characteristics of human visual perception and RGB channel information, as shown in formula (1). （1） in, For the pixels in the input image The corresponding brightness value, the pixel For any pixel in the input image, and pixels The number of rows and columns, , and These are the sensitivity coefficients of the human eye to red, green, and blue, respectively. , and pixels Normalized pixel values ​​in the red, green, and blue channels; Step A2: Calculate the global brightness of the input image based on the brightness values ​​of each pixel in the input image, as shown in formula (2). （2） in, To input the global brightness of the image, and These represent the total number of rows and columns of pixels in the input image, respectively.

3. The image adaptive enhancement method based on linear dynamic weight fusion according to claim 2, characterized in that: Step C involves calculating the local brightness standard deviation of each pixel in the input image and then calculating the global brightness variance based on the local brightness standard deviation. The specific steps are as follows. Step C1: Calculate and obtain the local brightness standard deviation of each pixel in the input image, as shown in formula (3). （3） in, For the pixels in the input image The local brightness standard deviation To input the pixels in the image Pixels in a 3x3 grid centered The brightness of the pixel To input the pixels in the image Any pixel in a 3×3 grid centered on the center and pixels The number of rows and columns, For all pixels The mean, For pixels The total number; Step C2: Calculate the global luminance variance based on the local luminance standard deviation, as shown in formula (4). （4） in, This represents the global brightness variance.

4. The image adaptive enhancement method based on linear dynamic weight fusion according to claim 3, characterized in that: Step D involves calculating the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, then calculating the weight values ​​of each pixel in the Gaussian kernel based on the Gaussian kernel standard deviation to obtain the spatial domain Gaussian kernel matrix. The specific steps are as follows. Step D1: Calculate the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, as shown in formula (5). （5） in, The standard deviation of the Gaussian kernel. and These are the maximum and minimum values ​​of the Gaussian kernel standard deviation, respectively. To adjust the coefficient, This represents the maximum global brightness variance. Step D2: Calculate the weight values ​​of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation and obtain the spatial domain Gaussian kernel matrix, as shown in formula (6). （6） in, For the Gaussian nucleus relative to the nucleus center pixels Weight value, the pixel For the Gaussian nucleus relative to the nucleus center The coordinates of any pixel, The coordinates are the center coordinates of the Gaussian kernel.

5. The image adaptive enhancement method based on linear dynamic weight fusion according to claim 4, characterized in that: Step E involves performing a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix. Then, the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix are multiplied element-wise in the frequency domain to obtain the frequency domain processing result. The specific steps are as follows. Step E1 involves performing a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix, as shown in formula (7). ； （7） in, This is the transformed color channel matrix. For frequency domain coordinates, The input image color channel matrix, the input image color channel matrix It includes a red channel matrix, a green channel matrix, and a blue channel matrix. The transformed spatial domain Gaussian kernel matrix, For the spatial domain Gaussian kernel matrix, The imaginary unit; Step E2 involves performing element-wise multiplication in the frequency domain on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result, as shown in formula (8). （8） in, This is the result of frequency domain processing.

6. The image adaptive enhancement method based on linear dynamic weight fusion according to claim 5, characterized in that: Step F involves performing a two-dimensional inverse fast Fourier transform on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component. Then, the global brightness standard deviation threshold is calculated using the local brightness standard deviation. The specific steps are as follows. Step F1 involves performing a two-dimensional inverse fast Fourier transform on the frequency domain processing result to map it back to the spatial domain and obtain the first image illumination component, as shown in formula (9). （9） in, This is the first image illumination component; Step F1: Calculate the global brightness standard deviation threshold using the local brightness standard deviation, as shown in formula (10). （10） in, This is the global brightness standard deviation threshold.

7. The image adaptive enhancement method based on linear dynamic weight fusion according to claim 6, characterized in that: Step G involves comparing the local brightness standard deviation of each pixel with the global brightness standard deviation threshold to obtain the standard deviation comparison result. Then, based on the standard deviation comparison result, the first image illumination component of each pixel is processed to obtain the second image illumination component. Specifically, if the local brightness standard deviation of a pixel is less than the global brightness standard deviation threshold, the second image illumination component of that pixel is equal to the first image illumination component. If the local brightness standard deviation of a pixel is greater than the global brightness standard deviation threshold, the second image illumination component of that pixel is half of the first image illumination component.

8. The image adaptive enhancement method based on linear dynamic weight fusion according to claim 7, characterized in that: Step H involves calculating the color channel reflectance component using the illumination components of the original and second input images, then normalizing the color channel reflectance component and obtaining the detail enhancement term. The specific steps are as follows. Step H1: Calculate the color channel reflectance component using the illumination components of the original input image and the second image, as shown in formula (11). （11） in, For the color channel reflection component, To reduce the ellipsis factor, if the illumination component of the second image is equal to the illumination component of the first image, then If the second image illumination component is 1, and the second image illumination component is half the first image illumination component, then It is 0.5; Step H2 involves normalizing the color channel reflection components and obtaining the detail enhancement term, as shown in formula (12). （12） in, This is for detail enhancements.

9. The image adaptive enhancement method based on linear dynamic weight fusion according to claim 8, characterized in that: Step I: Based on the brightness values ​​of each pixel in the input image, a detail enhancement term is used to enhance the image and obtain the enhanced image. Then, the enhanced image is output to complete the image enhancement operation, as shown in formula (13). （13） in, To enhance the post-image.

10. An image adaptive enhancement system based on linear dynamic weight fusion, wherein the specific enhancement process of the image adaptive enhancement system is based on the image adaptive enhancement method according to any one of claims 1-9, characterized in that: It includes a brightness calculation module, a brightness comparison module, a variance calculation module, a Gaussian kernel matrix acquisition module, a frequency domain processing module, an illumination component calculation module, an illumination component processing module, a detail enhancement item acquisition module, and an image enhancement module. The brightness calculation module is used to calculate the brightness value of each pixel in the input image using the characteristics of human visual vision and RGB channel information, and then calculate the global brightness of the input image based on the brightness value of each pixel in the input image. The brightness comparison module is used to compare the global brightness of the input image with a set brightness threshold. If the global brightness of the input image is lower than the set brightness threshold, image enhancement will continue. If the global brightness of the input image is higher than the set brightness threshold, the input image will be output directly. The variance calculation module is used to calculate and obtain the local brightness standard deviation of each pixel in the input image, and then calculate the global brightness variance based on the local brightness standard deviation. The Gaussian kernel matrix acquisition module is used to calculate the Gaussian kernel standard deviation using the local brightness standard deviation and the global brightness variance, and then calculate the weight value of each pixel in the Gaussian kernel of the input image based on the Gaussian kernel standard deviation to obtain the spatial domain Gaussian kernel matrix. The frequency domain processing module is used to perform a two-dimensional fast Fourier transform on the input image color channel matrix and spatial domain Gaussian kernel matrix to obtain the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix, and then perform element-wise frequency domain multiplication on the transformed color channel matrix and the transformed spatial domain Gaussian kernel matrix to obtain the frequency domain processing result. The illumination component calculation module is used to perform a two-dimensional inverse fast Fourier transform on the frequency domain processing results to map them back to the spatial domain and obtain the first image illumination component, and then use the local brightness standard deviation to calculate the global brightness standard deviation threshold. The illumination component processing module is used to compare the local brightness standard deviation of each pixel with the global brightness standard deviation threshold and obtain the standard deviation comparison result. Then, based on the standard deviation comparison result, the first image illumination component of each pixel is processed to obtain the second image illumination component. The detail enhancement module is used to calculate the color channel reflection component using the illumination components of the original image and the second image of the input image, and then normalize the color channel reflection component to obtain the detail enhancement item. The image enhancement module is used to enhance the image based on the brightness value of each pixel in the input image using detail enhancement terms, and then output the enhanced image to complete the image enhancement operation.

Images