Low-illuminance image enhancement method based on virtual exposure and dual fusion

Abstract

A low-illuminance image enhancement method based on virtual exposure and dual fusion, which relates to the technical field of image processing. The method comprises the following steps: step 1, receiving an original image; step 2, performing different levels of exposure on the original image to acquire a plurality of virtual images, and generating a virtual-image sequence; step 3, calculating a weight map for each virtual image in the virtual-image sequence, and obtaining a brightness-enhanced image after fusion; and step 4, extracting detail information from the brightness-enhanced image and fusing same with details of the original image, so as to obtain an enhanced image. The method has a relatively small brightness distortion and contrast distortion, and can effectively preserve visual information of an image itself, making it more compatible with the visual enjoyment of human eyes.

Description

Low illumination image enhancement method based on virtual exposure and double fusion Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a low-illumination image enhancement method based on virtual exposure and double fusion. Background Art

[0002] Digital image processing systems are widely used in industrial production, video surveillance, intelligent transportation, remote sensing monitoring, and other fields, playing a vital role in people's daily lives and the military. However, in poor lighting conditions, such as indoors, at night, and in overcast weather, the reflected light from objects is weak, resulting in image distortion and high noise levels, severely impacting image quality. In areas such as video surveillance, intelligent transportation, and autonomous driving, where accurate all-weather image information is required, uneven low-light conditions not only impair human visual perception but also pose significant challenges to subsequent machine systems in image analysis, image recognition, and target detection, potentially leading to system failure.

[0003] Traditional infrared fill light hardware solutions can obtain clear imaging in a completely dark environment, but can only form monochrome grayscale images, with serious color information loss and large noise. For example, the image enhancement model training, image enhancement method and electronic equipment disclosed in the publication number: CN115953309A.

[0004] In poor lighting conditions, camera-collected image information is severely lost, impacting both human visual perception and the performance of machine vision systems. Traditional image enhancement methods often suffer from over-enhancement and poor adaptability when processing images with uneven lighting. Therefore, using software to enhance low-light image detail and restore scene color information as closely as possible has become a pressing need across various fields.

[0005] Summary of the Invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a low-illumination image enhancement method based on virtual exposure and dual fusion.

[0007] The technical solution adopted by the present invention to solve the technical problem is: a low-light image enhancement method based on virtual exposure and dual fusion includes the following steps:

[0008] Step 1: Receive the original image;

[0009] Step 2: Expose the original image at different levels to obtain several virtual images and generate a virtual image sequence;

[0010] Step 3: Calculate the weight map of each virtual image in the virtual image sequence, and obtain the brightness enhanced image after fusion;

[0011] Step 4: Extract detail information from the brightness enhanced image and fuse it with the details of the original image to obtain an enhanced image.

[0012] The second step includes the following sub-steps:

[0013] 2-1: Construct a brightness reducer;

[0014] 2-2: Construct brightness enhancer;

[0015] 2-3: Based on the brightness reducer and brightness enhancer, different exposure processing is performed on the original image to obtain a virtual image sequence.

[0016] The brightness reducer in 2-1 is implemented using a gamma function;

[0017] The weakening function P is expressed as follows: P: {y = αx γ} (5)

[0018] Among them, x represents the input, y represents the output, α is the gain control factor, γ is the exponential control factor, and γ>1.

[0019] The brightness enhancer in 2-2 is implemented using a quadratic function; the enhancement function H is expressed by the following formula: H: {y = x + k × x (1-x)} (7)

[0020] Among them, x represents the input, y represents the output, and k represents the control coefficient.

[0021] The step three includes the following sub-steps:

[0022] 3-1: Calculate the contrast of a single virtual image;

[0023] 3-2: Calculate the saturation of a single virtual image;

[0024] 3-3: Calculate the weight of each virtual image in the virtual image sequence based on the contrast and saturation of the image;

[0025] 3-4: Perform image fusion based on the calculated weights to obtain a brightness enhanced image.

[0026] In step 3-1, the absolute value after Laplace filtering is used as the contrast factor, which includes the following sub-steps:

[0027] 3-1-1: Convert the virtual image into a grayscale image;

[0028] 3-1-2: Normalize pixel values ​​to the [0,1] range;

[0029] 3-1-3: Perform Laplace filtering on the virtual image. Its expression and filter template are: C k =I*h (12)

[0030] Where C represents contrast, I is the image whose contrast is to be determined, and h is the Laplace filter.

[0031] The saturation in step 3-2 is obtained by calculating the standard deviation of the three chromaticity channels, including the following sub-steps:

[0032] 3-2-1: Extract the R, G and B color components of the virtual image;

[0033] 3-2-2: Calculate the mean of the RGB color components of each pixel in the virtual image;

[0034] 3-2-3: Calculate the color standard deviation of a single virtual image to determine the saturation coefficient S ij,k , the specific calculation formula is as follows:

[0035] Where: I R , I G , I B are the pixel values ​​of the three color channels R, G, and B, and μ is the mean of the three.

[0036] The 3-3 step includes the following sub-steps:

[0037] 3-3-1: Multiply the two measurement factors to get the weight;

[0038] 3-3-2: Perform normalization processing to finally obtain the initial fusion weight W.

[0039] To avoid the halo effect caused by too fast weight changes, we use pyramids to decompose the image in the algorithm and achieve image fusion through multi-resolution. The above 3-4 includes the following sub-steps:

[0040] 3-4-1: Perform Laplacian pyramid decomposition on each virtual image and Gaussian pyramid decomposition on its weight map to obtain images and weight maps of different resolutions;

[0041] 3-4-2: At each layer, the weight of the weight map pyramid and the Laplace pyramid coefficient of the corresponding position are weighted and summed to obtain the fused new Laplace pyramid L{I F} l ;

[0042] 3-4-3: Laplace pyramid L{I F} l Reconstruct and finally get the fused image I F , which is the brightness enhanced image.

[0043] In the fourth step, the brightness enhanced image is converted to the HSV space, and the detail information of the brightness enhanced image is extracted and PCA fused with the details of the original image to finally obtain an image with enhanced details and brightness.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] (1) A unified framework for non-uniform illumination image enhancement based on virtual exposure and secondary image fusion is proposed, which fully utilizes the information in the original image and effectively improves the details of the enhanced image without causing color distortion.

[0046] (2) A strategy for generating a series of artificial multi-exposure images using a segmented mapping function is proposed. By combining a brightness enhancer and a brightness reducer, multi-exposure images are virtually generated, avoiding over-enhancement of local areas caused by uneven lighting.

[0047] (3) This method is fully parameter-adaptive and can take into account both processing speed and processing effect. It does not require large-scale data sets for learning and training and is suitable for low-light image enhancement with uneven illumination distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Fig. 1 is a flow chart of the present invention;

[0049] FIG2 is a graph showing a brightness reducer function;

[0050] FIG3 is a function curve diagram of a brightness enhancer. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0052] Referring to Figures 1-3, this application is a technology that fuses image sequences of different exposures acquired under the same scene, extracts clear information from each image, and synthesizes a new image. The working principle is expressed as follows: e =F{I i} (1)

[0053] Among them, I i is the i-th image to be fused in the image sequence, I e is the fused image, and F{} is the fusion function.

[0054] Specifically, the low-light image enhancement method based on virtual exposure and dual fusion includes the following steps:

[0055] Step 1: Receive the original image; the operator inputs a single low-light image.

[0056] Step 2: Expose the original image at different levels to obtain several virtual images and generate a virtual image sequence;

[0057] For the case of only one low-light image, in order to generate other image sequences for fusion, the simplest processing method is to use the tone mapping operator to directly process the original image so that I i ←I0. Assume that the original image is represented as I0, f i (·) is the image grayscale transformation function, then the processed image I i For: I i =f i (I0) (2)

[0058] Among them, f i It should satisfy the properties of monotonicity and boundedness:

[0059] Since the brightness transformation is not the result of real exposure, it is referred to as a virtual exposure device in this application. For a virtual exposure device, in order to obtain different brightness transformation functions, f i Considering that in addition to enhancing low-brightness areas during image processing, it may also be necessary to suppress over-enhanced image brightness areas. Therefore, we construct a brightness enhancer and a brightness reducer to achieve different exposures. It is assumed that the parameter values ​​required for image transformation are related to the overall brightness of the image scene and have a nonlinear relationship. The enhancement function and the reduction function are H and P, respectively, which are expressed as:

[0060] Among them, the number processed by the H function is NH, the number processed by the P function is NL, and the total number N=NH+NL.

[0061] In the step 2, the original image is processed based on the brightness enhancer and attenuator to obtain a virtual image sequence.

[0062] The second step includes the following sub-steps:

[0063] 2-1: Construct a brightness reducer;

[0064] 2-2: Construct brightness enhancer;

[0065] 2-3: Based on the brightness reducer and brightness enhancer, different exposure processing is performed on the original image to obtain a virtual image sequence.

[0066] Specifically, the brightness reducer in 2-1 is implemented using a gamma function;

[0067] Its attenuation function P is expressed as follows: P:{y=αx γ} (5)

[0068] Where x represents the input, y represents the output, α is the gain control factor, γ is the exponential control factor, and γ>1. Setting different γ can control the output of different images. The processed output image I i The expression is as follows: I i =P(I0,γ i ) (6)

[0069] i is a natural number. Its conversion curve is shown in Figure 2. In this embodiment, the brightness reducer uses gamma transformation to generate N L A virtual exposure image.

[0070] The brightness enhancer in 2-2 is implemented using a quadratic function. Compared to the gamma function, its enhancement of high-brightness areas can exceed the function's extreme value of 1, and is closer to the overexposure of bright objects. The input-output relationship of the virtual exposure enhancer set based on the image brightness control parameters, i.e., the enhancement function H, is expressed as follows: H: {y = x + k × x (1-x)} (7)

[0071] Among them, x represents the input, y represents the output, and k represents the control coefficient. By setting the value of the control parameter k, different images can be obtained. The image I after processing is i The expression is as follows: I i =H(I0,k i ) (8)

[0072] In formula (8), I0 represents the original low-light image, and each brightness value is in the range of [0,1].

[0073] Figure 3 shows the output curves for different k values. It can be seen that since the H function has a value greater than 1, combined with formula (7), the virtual image I i Each pixel value in can be expressed as:

[0074] Among them, k i is the exposure control parameter of the i-th image, and when it is brought into the virtual exposure enhancer, we get N H A virtual exposure-enhanced image.

[0075] As shown in Figure 3, as k increases, the slope of the transformation curve increases, and the image brightness is enhanced to a greater extent. This shows that the brightness of the generated image depends entirely on the parameters of the virtual exposure enhancer. By setting an appropriate k, the desired image brightness can be achieved.

[0076] In order to improve the adaptability of the algorithm in this embodiment, we use the grayscale average value of the virtual exposure image as the maximum grayscale expected value to calculate the k value. Assume that the grayscale average value of a virtual exposure image is μ k , the desired virtual exposure image grayscale average value is ξ, which will satisfy the exposure image grayscale average value μ k The k value that is closest to ξ is used as the control parameter to maximize the exposure. In addition, since the information content and exposure level of the original image are different, in order to constrain the degree of virtual exposure enhancement, the estimated The value is constrained and the maximum and minimum thresholds are set using k H and k L Indicates that The final estimated The value is expressed as:

[0077] and

[0078] In the actual application of this paper, we set the expected value of the average grayscale of the image ξ to 0.5 (the grayscale value range of the image is [0,1]), and set k H =12,k L =6. Among them, K L To prevent uneven illumination, the mean value reaches 0.5 and cannot be enhanced. H Prevents extreme low-light images from being over-enhanced.

[0079] According to formula (8), we can get the k value that makes the average grayscale value of the enhanced image ξ as the maximum exposure value. The number of virtual exposure sequence images N is set according to the need. H Parameter adjustment of virtual exposure device k i You can get N Himage sequences. The calculation formula is as follows: i =(i×k) / N H (11)

[0080] This paper uses a segmented virtual exposure function to perform virtual exposure on low-light images, which is equivalent to targeted processing of a certain area, so that the low-light area of ​​the original image can be brightened, while the high-exposure area can be better suppressed to obtain detailed information suitable for human eye observation.

[0081] Step 3: Calculate the weight map of each virtual image in the virtual image sequence, and obtain the brightness-enhanced image after fusion. Specifically, based on the image contrast and saturation information description factors, calculate and generate the weight map of each virtual image in the virtual image sequence, and realize multi-scale fusion of multi-exposure images to obtain the brightness-enhanced image.

[0082] To ensure that the fused image conforms to the characteristics of the human visual system, in these virtual exposure images, smooth areas caused by overexposure or underexposure should be given a smaller weight, while areas with good exposure and rich details should be assigned a larger weight. Based on this consideration, this application uses two measurement factors, contrast and saturation, to fuse and restore image information.

[0083] The step three includes the following sub-steps:

[0084] 3-1: Calculate the contrast of a single virtual image;

[0085] 3-2: Calculate the saturation of a single virtual image;

[0086] 3-3: Calculate the weight of each virtual image in the virtual image sequence based on the contrast and saturation of the image;

[0087] 3-4: Perform image fusion based on the calculated weights to obtain a brightness enhanced image.

[0088] Image Contrast (Contrast) C: Contrast represents the amount of image detail. The greater the contrast, the better the image detail is, and the easier it is for the human eye to distinguish. In step 3-1 of this application, the absolute value after Laplace filtering is used as the contrast factor. This term is mainly obtained by calculating the weight coefficient of the image edge change. It includes the following sub-steps:

[0089] 3-1-1: Convert the virtual image into a grayscale image;

[0090] 3-1-2: Normalize pixel values ​​to the [0,1] range;

[0091] 3-1-3: Perform Laplace filtering on the virtual image. Its expression and filter template are: C k =I*h (12)

[0092] Where C represents contrast, I is the image whose contrast is to be determined, and h is the Laplace filter.

[0093] The contrast calculated by this coefficient is mainly used to distinguish the degree of jump of the point relative to the surrounding area, so the absolute value of the pixel at this position is taken as the final contrast parameter.

[0094] Saturation (S): Saturation is an important indicator of image vividness, representing the vividness of the color. The higher the saturation, the more vivid the image. Saturation can be quantified by the RGB channels of each pixel in the image. In step 3-2, saturation is obtained by calculating the standard deviation of the three chromaticity channels, which includes the following sub-steps:

[0095] 3-2-1: Extract the R, G and B color components of the virtual image;

[0096] 3-2-2: Calculate the mean of the RGB color components of each pixel in the virtual image;

[0097] 3-2-3: Calculate the color standard deviation of a single virtual image to determine the saturation coefficient S ij,k , the specific calculation formula is as follows:

[0098] Where: I R , I G , I B are the pixel values ​​of the three color channels R, G, and B, and μ is the mean of the three.

[0099] Traditional image fusion algorithms do not consider the correlation between pixels, but instead directly fuse corresponding pixels in the source image to produce a new image. While these methods are simple and computationally inefficient, they cannot effectively manage the relationships between pixels in the source image, resulting in significant loss of texture detail. Therefore, it is necessary to improve and optimize existing spatial-domain fusion rules to enhance the quality of image fusion without significantly increasing computational complexity.

[0100] The 3-3 step includes the following sub-steps:

[0101] 3-3-1: Multiply the two measurement factors to get the weight;

[0102] 3-3-2: Perform normalization to ensure that the weight parameter is used for each pixel value, and finally obtain the initial fusion weight W.

[0103] The specific formula for fusing N virtually exposed image sequences is as follows:

[0104] Among them, C k and S k are the contrast and saturation of the k-th image, W k is the weight graph. c 、ω s The contrast factor C and saturation factor S are used to control the degree of influence of the weight map W. For each fusion source, the two measurement factors have the same influence on the generated weight map, so ω is taken in the algorithm. c =ω s =1.

[0105] In order to obtain consistent results, the weights of each exposure image are normalized, and the weight of the pixel point at (i, j) in the kth image among the N virtual exposure images is obtained as:

[0106] is the kth normalized weight graph, where in order to avoid the denominator being zero, an ε value is introduced, and ε=10 -12 .

[0107] The first row is the virtual exposure image sequence I k , the second row is the contrast weight C k , the third line is the saturation weight S k , the last line is the normalized weight W k The constructed normalized weight map can effectively outline the important parts of the image, effectively enhancing the unexposed areas in the low-light image while effectively preserving the well-exposed areas in the original image without introducing any artifacts.

[0108] The final enhanced image is obtained by fusing the multi-exposure image with the weight map. F It can be expressed as:

[0109] Among them, I k represents the kth input image, is the k-th normalized weight map.

[0110] To avoid the halo effect caused by too fast weight changes, we use pyramids to decompose the image in the algorithm and achieve image fusion through multi-resolution. The above 3-4 includes the following sub-steps:

[0111] 3-4-1: Perform Laplacian pyramid decomposition on each virtual image and Gaussian pyramid decomposition on its weight map to obtain images and weight maps of different resolutions. G / L represent Gaussian pyramid operation and Laplacian operation respectively. The Laplacian pyramid decomposition of the lth layer of image A is recorded as L{A} l , the image B is decomposed into the Gaussian pyramid layer l G{B} l ;

[0112] 3-4-2: Using formula (18), we perform weighted summation on each layer based on the weight of the weight map pyramid and the Laplace pyramid coefficient at the corresponding position to obtain the fused new Laplace pyramid L{I F} l The calculation formula is:

[0113] Where N represents the number of input images, Represents the normalized weight, l represents the number of layers of pyramid decomposition (0≤l≤M), and the default maximum number of layers is 5.

[0114] 3-4-3: Laplace pyramid L{I F} l Reconstruct and finally get the fused image I F , which is the brightness enhanced image. The fusion process is as follows.

[0115] Among them, ↑ d is the upsampling operator, d is the sampling factor, d = 2 l-1 .

[0116] Step 4: Extract detail information from the brightness enhanced image and fuse it with the details of the original image to obtain an enhanced image.

[0117] In the fourth step, the brightness enhanced image is converted to the HSV space, and the detail information of the brightness enhanced image is extracted and PCA fused with the details of the original image to finally obtain an image with enhanced details and brightness.

[0118] Specifically, the original low-light image contains valuable detail information. Extracting this detail information from the original image will be able to compensate for the detail information that may have been lost during the initial fusion process. Given that hue (H), saturation (S), and brightness (V) are independent of each other in the HSV color space, we convert the initially fused image from RGB space to HSV space, extract its V component information, and then fuse it with the grayscale detail information of the original image. The formula for converting RGB space to HSV space is:

[0119] Assume that a grayscale image consists of background information and detail information. In order to obtain the background information of the image, this paper designs a simple low-pass filter. It can be expressed as: V b =V*g (22)

[0120] Where g is a low-pass filter operator, and the filter size is set to 5.

[0121] Thus, we can get the detailed information of image V. That is, V h =VV b (twenty three)

[0122] Among them, V b represents the smoothing component; V h Represents the detail component, which consists of texture information and some image noise.

[0123] For the original image, its grayscale image transformation is as follows: I g =R*0.299+G*0.587+B*0.114 (24)

[0124] The same method as formula (22) and (23) can be used to obtain the detail information I of the original image grayscale image Ig h .

[0125] Since the overall information and structure of the images to be fused are similar, no registration is required. This application designs an image fusion algorithm based on principal component analysis (PCA) to determine the weighted coefficients to enhance the significant features of the image after the virtual exposure image is fused, that is, F = ω v V h +ω i I h (25)

[0126] Among them, F is the fused image, ω v 、ω i are the corresponding weighting coefficients respectively.

[0127] In this method, the eigenvectors and corresponding eigenvalues ​​of the source image are calculated to find the principal components of the approximate image, and the weights of the images to be fused are determined based on the principal components.

[0128] PCA fuses the images based on the correlation between several selected component images, compresses the data related to each other in the images, and can highlight the characteristics of the images.

[0129] This application is based on the fact that a single image can generate an image sequence to be fused through an attenuator and an enhancer, and an image with rich detail information is generated through two image fusions.

[0130] This method minimizes brightness and contrast distortion, effectively preserving the image's inherent visual information and aligning with human perception. The default basic parameters of this method are independent of external system inputs. The required transformation parameters are automatically calculated during algorithm implementation based on image content features. This method exhibits good adaptability and robustness, and can be applied to fields such as video surveillance and scene restoration.

[0131] The above descriptions are merely optional embodiments of the present invention and do not limit the patent scope of the present invention. All equivalent structural transformations made using the contents of the present description and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included in the patent protection scope of the present invention.

Claims

1. A low-light image enhancement method based on virtual exposure and dual fusion, characterized in that: The following steps are involved: Step 1: Receive the original image; Step 2: Expose the original image at different levels to obtain several virtual images and generate a virtual image sequence; Step 3: Calculate the weight map of each virtual image in the virtual image sequence, and obtain a brightness enhanced image after fusing each virtual image; Step 4: Extract detail information from the brightness enhanced image and fuse it with the details of the original image to obtain an enhanced image.

2. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 1, characterized in that: The second step includes the following sub-steps: 2-1: Construct a brightness reducer; 2-2: Construct brightness enhancer; 2-3: Based on the brightness reducer and brightness enhancer, different exposure processing is performed on the original image to obtain a virtual image sequence.

3. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 2, characterized in that: The brightness reducer in 2-1 is implemented using a gamma function; The attenuation function P is expressed as follows: P:{y=αx γ } (5) Among them, x represents the input, y represents the output, α is the gain control factor, γ is the exponential control factor, and γ>1.

4. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 2, characterized in that: The brightness enhancer in 2-2 is implemented using a quadratic function; The enhancement function H is expressed as follows: H:{y=x+k×x(1-x)} (7) Among them, x represents the input, y represents the output, and k represents the control coefficient.

5. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 1, characterized in that: The step three includes the following sub-steps: 3-1: Calculate the contrast of a single virtual image; 3-2: Calculate the saturation of a single virtual image; 3-3: Calculate the weight of each virtual image in the virtual image sequence based on the contrast and saturation of the image; 3-4: Perform image fusion based on the calculated weights to obtain a brightness enhanced image.

6. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 5, characterized in that: In 3-1, the absolute value after Laplace filtering is used as the contrast factor, which includes the following sub-steps: 3-1-1: Convert the virtual image into a grayscale image; 3-1-2: Normalize pixel values ​​to the [0,1] range; 3-1-3: Perform Laplace filtering on the virtual image. The expression and filter template are as follows: C k =I*h (12) Where C represents contrast, I is the image whose contrast is to be determined, and h is the Laplace filter.

7. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 5, characterized in that: The saturation in 3-2 is obtained by calculating the standard deviation of the three chromaticity channels, including the following sub-steps: 3-2-1: Extract the R, G and B color components of the virtual image; 3-2-2: Calculate the mean of the RGB color components of each pixel in the virtual image; 3-2-3: Calculate the color standard deviation of a single virtual image to determine the saturation coefficient S ij,k , the specific calculation formula is as follows: Where: I R , I G , I B are the pixel values ​​of the three color channels R, G, and B, and μ is the mean of the three.

8. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 5, characterized in that: The 3-3 step includes the following sub-steps: 3-3-1: Multiply the two measurement factors to get the weight; 3-3-2: Perform normalization processing to finally obtain the initial fusion weight W.

9. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 5, characterized in that: The steps 3-4 include the following sub-steps: 3-4-1: Perform Laplacian pyramid decomposition on each virtual image, and perform Gaussian pyramid decomposition on its weight map to obtain images and weight maps of different resolutions; 3-4-2: At each layer, the weight of the weight map pyramid and the Laplace pyramid coefficient of the corresponding position are weighted and summed to obtain the fused new Laplace pyramid L{I F } l ; 3-4-3: Laplace pyramid L{I F } l Reconstruct and finally get the fused image I F .

10. The low-light image enhancement method based on virtual exposure and dual fusion according to claim 1, characterized in that: In the fourth step, the brightness enhanced image is converted to the HSV space, and the detail information of the brightness enhanced image is extracted and PCA fused with the details of the original image to finally obtain an image with enhanced details and brightness.

Images