A multi-exposure image fusion method, system and storage medium

Abstract

The present invention discloses a multi-exposure image fusion method, system and storage medium, belonging to the technical field of image processing. First, an input image is input, and a good exposure estimation weight and a chromaticity weight are calculated. Then, the Gaussian pyramid of the comprehensive weight map is convolved with the Laplacian pyramid of the input image to obtain an initial fusion pyramid. Secondly, in the brightest and darkest regions of interest, a fast local Laplacian filter (FLLF) is used for local detail enhancement, and then the maximum value of the Laplacian coefficients is taken to obtain a Laplacian pyramid with the maximum coefficients, and the coefficients in the regions of interest are updated into the initial fusion pyramid. Finally, the final fusion image is reconstructed through the Laplacian pyramid and output. The present invention can perform appropriate enhancement without losing information in the brightest and darkest regions, ensure the naturalness of the light and shadow transition, and make the image details clearer.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and more particularly to a multi-exposure image fusion method, system and storage medium. Background Art

[0002] With the rapid development of the ultra-high definition digital industry and digital photography technology, people's pursuit of ultra-high definition and cool visual experience is more urgent. At present, the methods of multi-exposure image fusion mainly include: high dynamic range imaging (HDR), exposure image fusion algorithm and multi-exposure fusion method based on deep learning. Among them, the high dynamic range imaging method overcomes the defect of detail loss in the brighter and darker regions of overexposed and underexposed images. This method captures several images with different exposures of the same scene and reconstructs them into an HDR image. HDR imaging generally includes two main steps: HDR reconstruction and tone mapping. First, capture multiple low dynamic range images with different exposure levels in the same scene, then reconstruct the HDR image by inverting the camera response function, and finally, in order to display on ordinary devices, convert the HDR into an LDR image through tone mapping. Although the HDR imaging technology can restore the entire dynamic range of the scene and make all details visible in one image, when generating the HDR image, the response function of the camera and the imaging exposure parameters of the input image are required, and the tone mapping is very time-consuming, and the details in the brighter and darker areas will also be lost, which all limit the application scope of this method. The exposure image fusion algorithm measures the quality of each pixel according to weights such as good exposure, saturation, and contrast for a series of low dynamic range images of the same scene with different exposure degrees, and then selects "good" pixels from the image sequence and combines them into the final result. This method skips the step of calculating HDR and can immediately synthesize the images with multiple exposures into a high-quality, low dynamic range image for display.

[0003] In summary, the main challenge of the method based on high dynamic range imaging is the estimation of the camera response function (CRF), which is an ill-posed problem and requires additional information and constraints to break problems such as self-similarity and exponential ambiguity. Based on the multi-exposure image fusion method, the operation speed is often slow, the weighted mapping is larger than the noise, and if directly applied to fusion, it may lead to various artifacts, and problems such as lack of spatial neighborhood information, unnatural over-brightness, and local color distortion will occur. The exposure fusion algorithm based on deep learning only learns to fuse a fixed number of images, and the scale of the multi-exposure image dataset is too small to cover all real scenes, and the training process has poor flexibility.

[0004] Based on this, many researchers have studied multi-exposure fusion algorithms. For example, the multi-exposure image fusion algorithm based on pyramid decomposition first performs Laplacian pyramid transformation on the LDR image, and Gaussian pyramid transformation on the weight map composed of contrast, exposure rate, and saturation, and then combines them into a fused image. The image obtained by this method will lose the details and information of local areas in the brighter and darker parts. Another example is a multi-exposure image fusion algorithm based on structural block decomposition. Although it can maintain good global contrast, halos will appear in areas with large differences in image intensity. Recently, exposure fusion methods based on detail enhancement have been applied to enhance the details in the brightest and darkest areas. These methods introduce a detail extraction mechanism to enhance details. These methods not only enhance the details in the brightest and darkest areas, but also enhance other pixels of all input LDR images, which will lead to the complexity of the algorithm and, in most cases, over-enhancement of details. These exposure image fusion algorithms often have problems such as unnatural over-darkening and over-lightening, local color distortion, and failure to properly adapt to the overall brightness of the input image sequence.

[0005] Therefore, how to provide a multi-exposure fusion method, system, and storage medium based on weighted average adaptive factors and local detail enhancement is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a multi-exposure fusion method, system, and storage medium, based on weighted average adaptive factors and local detail enhancement, to solve the problems in the prior art such as failure to properly adapt to the overall brightness of the input image sequence, lack of spatial neighborhood information, and over-enhancement of details.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] On the one hand, the present invention provides a multi-exposure image fusion method, including the following steps:

[0009] S100: Input an image to obtain the Laplacian pyramid of the input image;

[0010] S200: Traverse the input image using a window, and expand boundary pixels around the traversed image to calculate the good exposure weight information and chromaticity weight information of the window center pixel;

[0011] S300: Obtain a comprehensive weight map according to the good exposure weight information and the chromaticity weight information, and establish a Gaussian pyramid corresponding to the comprehensive weight map;

[0012] S400: Perform convolution processing on the Gaussian pyramid of the comprehensive weight map and the Laplacian pyramid of the input image to obtain an initial fusion pyramid;

[0013] S500: Obtain the bright - interest region and the dark - interest region of the input image and perform enhancement processing to obtain the maximum value of the Laplacian coefficient, and obtain the Laplacian pyramid with the maximum coefficient according to the maximum value of the Laplacian coefficient;

[0014] S600: Update the Laplacian pyramid with the maximum coefficient into the initial fusion pyramid, form the final fused image and output it.

[0015] Preferably, the calculation of the good - exposure weight information of the center pixel of the window in S200 includes:

[0016] S210: Calculate the sequence I of the input image:

[0017] I = {I1, I2, I3,....., I N}

[0018] In the formula, I represents the sequence of input images, N represents the number of images;

[0019] S211: Determine the weighted - average adaptive factor γ according to the sequence I, and the specific expression is:

[0020]

[0021] In the formula, I k (x, y) represents the pixel value of the k - th input image at (x, y), N is the number of input - image sequences, h and w are the sizes of the input image, and P(x, y) is the expected brightness;

[0022] S212: Divide the determined exposure - weight function into two parts, namely W ED and W EB , where the selection of the exposure - weight function is determined by the average brightness of the current pixel and the expected brightness P, and the specific expression is:

[0023]

[0024] S213: When is less than P(x, y), the weight function is W ED :

[0025]

[0026] S214: When is greater than P(x, y), the weight function is W EB , and the specific calculation process is:

[0027]

[0028] Preferably, the calculation formula for the desired brightness P(x, y) is as follows:

[0029]

[0030] where β is a parameter adjustable by the user and is used to control the proportion of the brightest and darkest pixel values, and I D and I B are respectively the darkest and brightest pixel values of the N input image sequences at the (x, y) position.

[0031] Preferably, the calculation of the chromaticity weight information of the central pixel of the S200 calculation window includes:

[0032] Calculating the chromaticity weight of the input image position:

[0033]

[0034] where C is the R, G, B three channels, is the pixel value of the R, G, B three channels, is the chromaticity weight at the (x, y) position of the k-th input image.

[0035] Preferably, the Gaussian pyramid of the S300 to obtain the comprehensive weight map includes calculating the initial fusion weight according to the chromaticity weight of the position:

[0036]

[0037] where ε is a very small coefficient, set to 10 -12 , is the good exposure weight function. When is less than P(x, y), otherwise

[0038] Preferably, the initial fusion pyramid of the S400 includes obtaining the initial fusion image using the Gaussian pyramid and the Laplacian pyramid:

[0039]

[0040] where l represents the number of layers of the pyramid, G{W k} l represents the Gaussian pyramid of the weight, L{W k} l represents the Laplacian pyramid of the input image, L{F First} l represents the Laplacian pyramid of the initial fusion image.

[0041] Preferably, the S500 for obtaining the maximum coefficient Laplacian pyramid includes:

[0042] S510: Calculate the average luminance of each bit (x, y) to determine the brightest and darkest regions of enhancement (ROE).

[0043]

[0044] In the formula, when it indicates that the pixel (x, y) is in the brightest region B(x, y), and when it indicates that the pixel (x, y) is in the darkest region D(x, y);

[0045] S520: Use the fast local Laplacian pyramid (FLLF) to enhance the details of the ROE region:

[0046]

[0047] In the formula, represents the k-th l-layer Laplacian pyramid of the enhanced ROE region;

[0048] S530: Perform a maximum operation on to obtain the image details:

[0049]

[0050] where represents the maximum Laplacian coefficient of the l-layer Laplacian pyramid of the enhanced region at the position (x, y);

[0051] S540: Obtain the maximum coefficient Laplacian pyramid according to the maximum value of the Laplacian coefficient.

[0052] Preferably, the S600 includes: updating the maximum value of the Laplacian coefficient into the Laplacian pyramid of the region of interest of the initial fused image to obtain the Laplacian pyramid of the final fused image, and the calculation formula is:

[0053]

[0054] In the formula, ROE(x, y) includes the normalized pixel values of the brightest, darkest, and normal regions. Among them, the pixel value of the brightest region is 1; the pixel value of the darkest region is 0; the pixel value of the normal region is 0.5.

[0055] On the other hand, the present invention provides a multi-exposure image fusion system, including:

[0056] An image input module, configured to input an image and obtain the Laplacian pyramid of the input image;

[0057] A preprocessing module, connected to the image input module, is configured to traverse the input image by using a window, expand boundary pixels around the traversed image, and calculate good exposure weight information and chromaticity weight information of the window center pixel;

[0058] An integration module, connected to the preprocessing module, is configured to obtain an integrated weight map according to the good exposure weight information and the chromaticity weight information, and obtain a Gaussian pyramid of the integrated weight map;

[0059] A fusion module, connected to the image input module and the integration module, is configured to convolve the Gaussian pyramid of the integrated weight map with the Laplacian pyramid of the input image to obtain an initial fusion pyramid;

[0060] A calculation module, connected to the fusion module, obtains regions of interest in the brightness and darkness of the input image, obtains the maximum value of the Laplacian coefficient, and obtains a Laplacian pyramid with the maximum coefficient according to the maximum value of the Laplacian coefficient;

[0061] An output module, connected to the calculation module, updates the Laplacian pyramid with the maximum coefficient into the initial fusion pyramid, constructs a final fused image and outputs it.

[0062] On the other hand, the present invention also provides a computer-readable storage medium, on which a computer program is stored, and characterized in that when the program is executed by a processor, the multi-exposure image fusion method as described above is implemented.

[0063] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a multi-exposure image fusion method, system and storage medium. Aiming at the problems existing in the existing solutions that the overall brightness of the input image sequence is not well adapted, the lack of spatial neighborhood information, excessive detail enhancement, etc., through exposure fusion of a weighted average adaptive factor, and for the problems of detail loss and excessive detail enhancement in the brightest and darkest regions, a local detail enhancement scheme is adopted. The present invention can perform appropriate enhancement without losing information in the brightest and darkest regions, ensure the naturalness of the light and shadow transition, and make the image details clearer. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to the provided drawings without creative efforts.

[0065] Figure 1Schematic flowchart of the multi-exposure image fusion method provided by the present invention;

[0066] Figure 2 Schematic structural diagram of the multi-exposure image fusion method system provided by the embodiment of the present invention. Specific embodiments

[0067] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0068] See the attached Figure 1 As shown, the embodiment of the present invention discloses a multi-exposure image fusion method, including the following steps:

[0069] S100: Input an image to obtain the Laplacian pyramid of the input image;

[0070] S200: Traverse the input image using a window, expand the boundary pixels around the traversed image, and calculate the good exposure weight information and chromaticity weight information of the window center pixel;

[0071] S300: Obtain a comprehensive weight map according to the good exposure weight information and chromaticity weight information, and establish a Gaussian pyramid corresponding to the comprehensive weight map;

[0072] S400: Perform convolution processing on the Gaussian pyramid of the comprehensive weight map and the Laplacian pyramid of the input image to obtain an initial fusion pyramid;

[0073] S500: Obtain the bright region of interest and the dark region of interest of the input image and perform enhancement processing to obtain the maximum Laplacian coefficient. Obtain the maximum coefficient Laplacian pyramid according to the maximum Laplacian coefficient;

[0074] S600: Update the maximum coefficient Laplacian pyramid into the initial fusion pyramid to form and output the final fusion image.

[0075] More specifically, first, an input image is taken. In the window area, the weight information of the window center pixel is calculated according to the weighted average adaptive exposure estimation weight algorithm and chromaticity weight estimation algorithm proposed in this paper. Then, the Gaussian pyramid of the comprehensive weight map is convolved with the Laplacian pyramid of the input image to obtain an initial fusion pyramid. Secondly, in the brightest and darkest regions of interest obtained, the fast local Laplacian filter (FLLF) is used for local detail enhancement, and then the maximum value of the Laplacian coefficients is taken to obtain the Laplacian pyramid with the maximum coefficients, and the coefficients in the regions of interest are updated to the initial fusion pyramid. Finally, the final fusion image is reconstructed through the Laplacian pyramid.

[0076] In a specific embodiment, calculating the well-exposed estimation weight information includes:

[0077] Generally, in multi-exposure images, the intensity values of the under-exposed or over-exposed partial regions of the images are close to 1 and 0 (corresponding to the darkest and brightest regions respectively), and most details will be lost in these regions, while the intensity values of the well-exposed regions are close to 0.5 and contain more image information. If higher weights are assigned to the pixel values near 0.5, the consistency with the input image in brightness distribution cannot be guaranteed. This requires the relative brightness of the expected values of different pixels to be consistent with the values in the input image, that is, the expected value of the brighter pixel should be larger, and the expected value of the darker pixel should be smaller. Therefore, an expected brightness P is needed to adapt to the brightness distribution of the input image sequence, and this expected brightness is determined by the weighted average of 0.5, the brightest and darkest pixel values.

[0078] To have better adaptability to the brightness and darkness of the input image, the present invention uses a weighted average adaptive factor, which is determined by the average brightness of all input images and the expected brightness value. According to the average value of the current pixel values of the input image sequence, the weights of the darker and brighter regions are adaptively adjusted to obtain better sensory effects and more image details.

[0079] The specific method of the weighted average adaptive algorithm for exposure evaluation weights is as follows:

[0080] For all input image sequences,

[0081] I = {I1, I2, I3,....., I N}

[0082] where I represents the sequence of input images, and N represents the number of images. First, calculate the average brightness and expected brightness of I, and thereby determine the weighted average adaptive factor γ, which determines the quality of the current pixel exposure.

[0083]

[0084] where Ik (x, y) represents the pixel value of the k-th input image at (x, y). N is the number of input image sequences, h and w are the sizes (number of rows and columns) of the input images, and P(x, y) is the desired brightness, which is determined by 0.5 and the brightest and darkest pixel values and is defined as follows:

[0085]

[0086] In the formula, β is a user-adjustable parameter used to control the proportion of the brightest and darkest pixel values. ID and IB are the darkest and brightest pixel values of the N input image sequences at the (x, y) position.

[0087] I D = Min(I1(x, y), I2(x, y), I3(x, y), ……, I N (x, y))

[0088] I B = Max(I1(x, y), I2(x, y), I3(x, y), ……, I N (x, y))

[0089] The exposure weight function determined by the weighted average adaptive factor γ is divided into two parts, namely W ED and W EB . The selection of the exposure weight function is determined by the average brightness of the current pixel and the desired brightness P.

[0090]

[0091] When is less than P(x, y), it means that the overall current pixel position is darker. At this time, the weight of the dark part is appropriately increased and the weight of the bright part is appropriately decreased. The weight function is W ED :

[0092]

[0093] When is greater than P(x, y), it means that the overall current pixel position is brighter. At this time, the weight of the bright part is appropriately increased and the weight of the dark part is appropriately decreased. The weight function is W EB :

[0094]

[0095] In a specific embodiment, calculating the chromaticity weight specifically includes:

[0096] Vivid colors often give greater visual impact. Therefore, the higher the saturation of the color, the greater the weight needs to be given. The specific algorithm is as follows:

[0097]

[0098] Among them, C represents the R, G, and B channels. is the chromaticity weight at the position (x, y) of the k-th input image. Then, the initial fusion weight is calculated, and the specific algorithm is as follows:

[0099]

[0100] Among them, ε is a very small coefficient. To avoid the case of a zero denominator, ε can be set to 10 -12 . is the well-exposed weight function. When is less than P(x, y), Otherwise

[0101] To avoid oversharpening of the weight map, a Gaussian filter is used to smooth the weight map. To avoid the problem of seams, a Gaussian pyramid and a Laplacian pyramid are used to obtain the initial fused image;

[0102]

[0103] l represents the number of layers of the pyramid, G{W k} l represents the Gaussian pyramid of the weight, L{W k} l represents the Laplacian pyramid of the input image, L{F First} l represents the Laplacian pyramid of the initial fused image.

[0104] In a specific embodiment, local detail enhancement includes:

[0105] When using a multi-resolution strategy to fuse multi-exposure images, although it retains sufficient details in the most normal regions, it will lose the fine details in the brightest and darkest regions. In this paper, a fast local Laplacian pyramid enhancement scheme is proposed to enhance the details in the brightest and darkest regions. First, we determine the brightest and darkest enhancement regions (ROE) by calculating the average brightness of each pixel (x, y).

[0106]

[0107] When , it means that the pixel (x, y) is in the brightest region B(x, y). When , it means that the pixel (x, y) is in the darkest region D(x, y). The details of the ROE region are enhanced using the fast local Laplacian pyramid (FLLF).

[0108] The larger the enhanced Laplacian pyramid coefficient, the richer the details contained in the current position (x, y), and for Take the maximum operation to obtain richer image details.

[0109]

[0110] Where represents the maximum Laplacian coefficient at the position (x, y) of the l-th layer Laplacian pyramid in the enhanced region. Then update this coefficient to the Laplacian pyramid of the region of interest of the initial fused image to obtain the Laplacian pyramid of the final fused image.

[0111]

[0112] In the formula, ROE(x, y) includes the pixel values of the brightest, darkest, and normal regions after normalization. Among them, the pixel value of the brightest region is 1; the pixel value of the darkest region is 0; the pixel value of the normal region is 0.5.

[0113] The final fused image is reconstructed through the Laplacian pyramid and output.

[0114] On the other hand, referring to the appendix Figure 2 As shown, this embodiment also provides a multi-exposure image fusion system, including:

[0115] An image input module for inputting an image to obtain the Laplacian pyramid of the input image;

[0116] A preprocessing module connected to the image input module, which is used to traverse the input image by a window, expand the boundary pixels around the traversed image, and calculate the good exposure weight information and chromaticity weight information of the window center pixel;

[0117] A comprehensive module connected to the preprocessing module, which is used to obtain a comprehensive weight map according to the good exposure weight information and the chromaticity weight information, and obtain the Gaussian pyramid of the comprehensive weight map;

[0118] A fusion module connected to the image input module and the comprehensive module, which is used to convolve the Gaussian pyramid of the comprehensive weight map with the Laplacian pyramid of the input image to obtain an initial fusion pyramid;

[0119] A calculation module connected to the fusion module, which acquires the regions of interest in the brightness and darkness of the input image, obtains the maximum Laplacian coefficient, and obtains the Laplacian pyramid with the maximum coefficient according to the maximum Laplacian coefficient;

[0120] An output module, connected to the calculation module, updates the maximum coefficient Laplacian pyramid into the initial fusion pyramid, forms a final fused image and outputs it.

[0121] In another aspect, the present embodiment also provides a computer-readable storage medium, on which a computer program is stored, characterized in that when the program is executed by a processor, the multi-exposure image fusion method as described above is implemented.

[0122] As can be seen from the above technical solutions, compared with the prior art, a multi-exposure image fusion method, system and storage medium disclosed by the present invention address the problems existing in the existing solutions, such as the lack of adaptation to the overall brightness of the input image sequence, the lack of spatial neighborhood information, and excessive detail enhancement. Through exposure fusion with a weighted average adaptive factor and a local detail enhancement scheme for problems such as detail loss and excessive detail enhancement in the brightest and darkest regions, the present invention can appropriately enhance while ensuring that information is not lost in the brightest and darkest regions, ensuring the naturalness of the light and shadow transition and making the image details clearer. The specific technical effects are as follows:

[0123] (1) The present invention takes into account the spatial neighborhood information of the image to ensure the integrity of the image information;

[0124] (2) The light and dark transitions are natural and there is no halation;

[0125] (3) The details in the darker and brighter regions are richer;

[0126] (4) It can maintain consistency with the input image in terms of brightness distribution;

[0127] (5) Through the algorithm of obtaining the good exposure weight of the image with a weighted average adaptive factor, and local detail enhancement in the brighter and darker regions, the image details are enriched;

[0128] (6) In cooperation with the chromaticity weight, a high-quality fused image is generated, which has most of the details of the original input image sequence while enhancing the details in the darker and brighter regions, and also ensures the naturalness of the light and shadow transition.

[0129] In this specification, each embodiment is described in a progressive manner. The key points of each embodiment are the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0130] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-exposure image fusion method, characterized in that, It includes the following steps: S100: Input an image to obtain the Laplacian pyramid of the input image; S200: Traverse the input image using a window, expand the boundary pixels around the traversed image, and calculate the good exposure weight information and chromaticity weight information of the window center pixel; S300: Obtain a comprehensive weight map based on the good exposure weight information and the chromaticity weight information, and build a Gaussian pyramid corresponding to the comprehensive weight map; S400: Perform convolution processing on the Gaussian pyramid of the comprehensive weight map and the Laplacian pyramid of the input image to obtain an initial fusion pyramid; S500: Obtain the bright and dark regions of interest in the input image and perform enhancement processing to obtain the maximum Laplacian coefficient. Obtain the Laplacian pyramid with the maximum coefficient based on the maximum Laplacian coefficient; S600: Update the Laplacian pyramid with the maximum coefficient to the initial fusion pyramid to form and output the final fused image; Among them, the calculation of the good exposure weight information of the window center pixel in S200 includes: S210: Calculate the sequence I of the input image: I = {I1, I2, I3,......, I N} In the formula, I represents the sequence of the input image, and N represents the number of images; S211: Determine the weighted average adaptive factor γ according to the sequence I. The specific expression is: where I k (x, y) represents the pixel value of the k-th input image at (x, y), N is the number of input image sequences, h and w are the sizes of the input images, and P(x, y) is the desired brightness; S212: Divide the determined exposure weight function into two parts according to the weighted average adaptive factor γ, namely W ED and W EB , where the selection of the exposure weight function is determined by the current pixel average brightness and the expected brightness P. The specific expression is as follows: S213: When is less than P(x, y), the weight function is W ED : S214: When is greater than P(x, y), the weight function is W EB , and the specific calculation process is as follows: The calculation of the chromaticity weight information of the window center pixel in S200 includes: Calculate the chromaticity weight at the position of the input image: where C represents the R, G, and B channels, are the pixel values of the R, G, and B channels, is the chromaticity weight at the position (x, y) of the k-th input image.

2. The multi-exposure image fusion method according to claim 1, wherein The calculation formula of the expected brightness P(x, y) is: where β is a parameter adjustable by the user and is used to control the proportion of the brightest and darkest pixel values, and I D and I B are respectively the darkest and brightest pixel values of the N-frame input image sequence at the (x, y) position.

3. A multi-exposure image fusion method according to claim 1, characterized in that, Obtaining the Gaussian pyramid of the comprehensive weight map in S300 includes calculating the initial fusion weight according to the chromaticity weight at the position: where ε is a calculation coefficient, is a good exposure weight function. When is less than P(x, y), otherwise 4. A multi-exposure image fusion method according to claim 3, wherein The initial fusion pyramid in S400 includes obtaining the initial fused image using the Gaussian pyramid and the Laplacian pyramid: where l represents the number of layers of the pyramid, G{W k} l represents the Gaussian pyramid of weights, L{W k} l represents the Laplacian pyramid of the input image, L{F First} l represents the Laplacian pyramid of the initial fused image.

5. A multi-exposure image fusion method according to claim 1, characterized in that Obtaining the Laplacian pyramid with the maximum coefficient in S500 includes: S510: Calculate the average brightness of each position (x, y) to determine the brightest and darkest enhanced regions ROE; In the formula, when , it indicates that the pixel (x, y) is in the brightest region B(x, y), and when , it indicates that the pixel (x, y) is in the darkest region D(x, y); S520: Use the fast local Laplacian pyramid FLLF to enhance the details of the ROE region; In the formula, represents the k-th Laplacian pyramid of the l-th layer in the enhanced ROE region; S530: For Take the maximum operation to obtain image details: where represents the largest Laplacian coefficient at the position (x, y) of the l-th layer Laplacian pyramid in the enhanced region; S540: Obtain the Laplacian pyramid with the maximum coefficient according to the maximum Laplacian coefficient.

6. A multi-exposure image fusion method according to claim 5, characterized in that, S600 includes: Update the maximum Laplacian coefficient to the Laplacian pyramid of the region of interest in the initial fused image to obtain the Laplacian pyramid of the final fused image. The calculation formula is: In the formula, ROE(x, y) includes the normalized pixel values of the brightest, darkest, and normal regions.

7. A multi-exposure image fusion system, characterized in that It includes: An image input module for inputting an image to obtain the Laplacian pyramid of the input image; A preprocessing module connected to the image input module for traversing the input image using a window, expanding the boundary pixels around the traversed image, and calculating the good exposure weight information and chromaticity weight information of the window center pixel; A comprehensive module connected to the preprocessing module for obtaining a comprehensive weight map based on the good exposure weight information and the chromaticity weight information, and obtaining the Gaussian pyramid of the comprehensive weight map; A fusion module, connected to the image input module and the synthesis module, for convolving the Gaussian pyramid of the synthesis weight map with the Laplacian pyramid of the input image to obtain an initial fusion pyramid; A calculation module, connected to the fusion module, to obtain the regions of interest in the brightness and darkness of the input image, obtain the maximum Laplacian coefficient, and obtain the Laplacian pyramid with the maximum coefficient according to the maximum Laplacian coefficient; An output module, connected to the calculation module, to update the Laplacian pyramid with the maximum coefficient into the initial fusion pyramid, form a final fused image and output it; Among them, calculating the good exposure weight information of the central pixel of the calculation window includes: S210: Calculate the sequence I of the input image: I = {I1, I2, I3,......, I N} In the formula, I represents the sequence of the input image, and N represents the number of images; S211: Determine the weighted average adaptive factor γ according to the sequence I, and the specific expression is: where I k (x, y) represents the pixel value of the k-th input image at (x, y), N is the number of input image sequences, h and w are the sizes of the input images, and P(x, y) is the expected luminance; S212: Divide the determined exposure weight function into two parts according to the weighted average adaptive factor γ, namely W ED and W EB , where the selection of the exposure weight function is determined by the current pixel average brightness and the expected brightness P, and the specific expression is: S213: When is less than P(x, y), the weight function is W ED : S214: When is greater than P(x, y), the weight function is W EB , and the specific calculation process is as follows: Calculating the chromaticity weight information of the central pixel of the calculation window includes: Calculating the chromaticity weight at the position of the input image: where C represents the R, G, and B channels, are the pixel values of the R, G, and B channels, is the chromaticity weight at the (x, y) position of the k-th input image.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the multi-exposure image fusion method according to any one of claims 1-6.

Images