A method and device for processing image wide dynamic range based on exposure fusion

Abstract

The present invention relates to the field of image processing technology, and specifically to a method and device for processing an image with a wide dynamic range based on exposure fusion, comprising the following steps: collecting multiple images of the same scene with different exposure ratios taken by the same camera, and fitting the image pixel values ​​to obtain a camera response model; obtaining a luminance map of the original input image, and obtaining an illumination map corresponding to the input image after downsampling and processing; performing adaptive histogram equalization on the input image with contrast restricted to obtain an image with enhanced contrast of the original image; obtaining a weight matrix based on the illumination map, and performing exposure fusion on the image, and performing linear stretching on the image to obtain an image processed with a wide dynamic range. The present invention artificially constructs images with different exposure ratios for exposure fusion, and can adaptively process different brightness areas, effectively enhancing the image's detailed texture and improving the quality of the dark areas of the image. Furthermore, the present invention requires low computing resources and can meet hardware real-time processing requirements.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method and device for processing an image with a wide dynamic range based on exposure fusion. Background Art

[0002] The method of using a global tone mapping curve for wide dynamic range image processing cannot be applied to pixels in all areas of the image. Therefore, processing operations such as gamma correction, histogram equalization, and segmented transformation are often used to adjust the overall contrast of the image. However, these adjustments have limited effects. Although they can improve the brightness and detail information of a certain area of ​​the image, they will sacrifice the quality of other local areas.

[0003] The method of using filter layering for wide dynamic range image processing has high computational complexity and is prone to haloing, layering, gradient reversal, and edge blurring. In addition, after the image is decomposed into multiple layers, it will occupy a large amount of resources, further limiting hardware performance and making it impossible to handle more complex filtering operations.

[0004] Therefore, this paper proposes an image wide dynamic range processing method and device based on exposure fusion. By artificially constructing images with different exposure ratios for exposure fusion, it can adaptively process different brightness areas, effectively enhance the image's detail texture, and improve the quality of the image's dark areas. It also requires low computing resources and can meet the hardware's real-time processing requirements. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention discloses a method and device for processing wide dynamic range images based on exposure fusion, which is used to solve the problem of low regional pixel quality caused by insufficient dynamic range of a single shot image, making the dark details in the shot image more obvious; at the same time, it reduces the computing resource and storage resource consumption of the wide dynamic range processing algorithm, enabling the algorithm to process real-time scenes in a hardware environment.

[0006] The present invention is achieved through the following technical solutions:

[0007] In a first aspect, the present invention provides an image wide dynamic range processing method based on exposure fusion, comprising the following steps:

[0008] S1 collects multiple images of the same scene with different exposure ratios taken by the same camera, and fits the image pixel values ​​to obtain a camera response model;

[0009] S2 inputs the input image into the camera response model to obtain an additional generated image, and obtains the brightness map of the original input image, and obtains the illumination map corresponding to the input image after downsampling and processing;

[0010] S3 performs contrast-limited adaptive histogram equalization on the input image to obtain an image with enhanced contrast of the original image;

[0011] S4 obtains a weight matrix according to the illumination map, and performs exposure fusion on the additional image obtained in S2 and the image in S3. The image is linearly stretched to obtain an image processed with a wide dynamic range.

[0012] Furthermore, in the method, a mathematical model is used to construct a camera response model, pixel values ​​of the image are extracted to fit the function model, and the generated images with different exposure ratios are compared with the real images with different exposure ratios taken, and the histogram distance between the generated images and the real images is measured. The camera response model sought minimizes the histogram distance between the two images.

[0013] Furthermore, in the method, a beta-gamma camera response model is selected, and the calculation formula for generating images with different exposure ratios is as follows:

[0014]

[0015] Where g(g) represents the camera response model, β and γ are two parameters of the model, e is the natural coefficient, P is the original image, and images with different exposure ratios are calculated using camera parameters a, b and exposure ratio k.

[0016] Furthermore, we set the exposure ratio k, select one of the photos, and continuously adjust the values ​​of parameters a and b in the camera response model so that the histograms of the generated images with different exposure ratios are closest to the histograms of the corresponding exposure ratio images taken in real life. The distance is measured using the Euclidean distance, which is calculated as follows:

[0017] Where D(Q,M) represents the histogram distance between the generated image Q and the real image M, 0≤q i ≤1 and 0≤m i ≤1 is the frequency corresponding to the pixel value i. For RGB three-channel images, the distance of the three channels is calculated separately, and the final distance is the mean of the three-channel histogram distance.

[0018] Among them, D R 、D G 、D B Represent the histogram distances of red, green and blue channels respectively.

[0019] Furthermore, in the method, the exposure ratio of the generated image is set to control the degree of wide dynamic range enhancement, wherein the generated image is consistent with the input image; the input image is sent to the camera response model for mapping calculation to obtain the generated image corresponding to the exposure ratio.

[0020] Furthermore, in the method, the input image of size is converted from RGB color space to HSV color space, and the three channels of red, green and blue are converted into three channels of hue, saturation and lightness, and the lightness channel is extracted as the lightness map of the image; the obtained lightness map of size is downsampled by times to obtain a lightness map of resolution, and the lightness map is smoothed edge-preservingly using bilateral filtering with a radius, and then a times upsampling interpolation operation is performed to restore the original image size to obtain the final image illumination map.

[0021] Furthermore, in the method, the weight matrix W required for exposure fusion is obtained according to the illumination map L, and the weight matrix is: W = L μ , where μ controls the degree of enhancement in different brightness intervals;

[0022] The images P′ and Q are exposed and fused, and the weight matrix W is used as the fusion weight coefficient of the pixels of the image P′, and the weight matrix 1-W is used as the fusion weight coefficient of the image Q. The fused image T is obtained, and its fused pixel value calculation formula is as follows:

[0023] T(x,y,c)=W(x,y,c)×P′(x,y,c)+(1-W(x,y,c))×Q(x,y,c),c∈R,G,B

[0024] Among them, x is the horizontal coordinate value of the pixel, y is the vertical coordinate value of the pixel, and c is the corresponding color channel. The image exposure fusion process is applied to the R, G, and B channels. The three channels are weightedly fused separately. The fusion process does not affect each other and is finally merged into an RGB image.

[0025] Furthermore, in the method, the fused image is linearly stretched, the maximum value of the RGB three channels of the fused image T is kept unchanged, the minimum value is mapped to the original input image P, and the three-channel linear stretching operation is performed on the middle pixels to obtain the final wide dynamic range processed image Z.

[0026] In a second aspect, the present invention provides an image wide dynamic range processing device based on exposure fusion, characterized by comprising:

[0027] The image dataset collection module is used to collect images of the same scene with different exposure ratios taken by the same camera, and store the captured images and their corresponding exposure compensation information;

[0028] The camera response model fitting module is used to construct a mathematical model to express the camera response model, and then fit the collected images with different exposure ratios taken in real scenes;

[0029] An image wide dynamic range processing module includes a memory and a processor, wherein the memory is used to store computer programs, processor input and output data, and related intermediate results. When the processor executes the computer program stored in the memory, the processor executes the image wide dynamic range processing method based on exposure fusion described in the first aspect.

[0030] The beneficial effects of the present invention are:

[0031] The present invention generates images with different exposure ratios for exposure fusion by constructing a camera response model, without the need for images with different exposure ratios shot in real scenes. Only one low dynamic range image is needed for wide dynamic range processing, without the need for multiple shooting operations, thus reducing the difficulty of shooting.

[0032] The present invention does not need to consider factors such as lens jitter and ghosting caused by shooting images with different exposure ratios. It can perform adaptive wide dynamic range processing for different brightness areas, effectively enhance the image's detail texture, and improve the quality of the image's dark areas. It also requires low computing resources and can meet the hardware's real-time processing requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0034] Figure 1 This is a principle and steps diagram of an image wide dynamic range processing method based on exposure fusion;

[0035] Figure 2 It is a network block diagram of a method for processing a single input image with a wide dynamic range. DETAILED DESCRIPTION

[0036] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0037] Example 1

[0038] Reference Figure 1As shown, this embodiment provides an image wide dynamic range processing method based on exposure fusion, including the following steps:

[0039] S1 collects multiple images of the same scene with different exposure ratios taken by the same camera, and fits the image pixel values ​​to obtain a camera response model;

[0040] S2 inputs the input image into the camera response model to obtain an additional generated image, and obtains the brightness map of the original input image, and obtains the illumination map corresponding to the input image after downsampling and processing;

[0041] S3 performs contrast-limited adaptive histogram equalization on the input image to obtain an image with enhanced contrast of the original image;

[0042] S4 obtains a weight matrix according to the illumination map, and performs exposure fusion on the additional image obtained in S2 and the image in S3. The image is linearly stretched to obtain an image processed with a wide dynamic range.

[0043] This embodiment constructs a camera response model required to generate different exposure ratios, including: using a commonly used mathematical model to construct a form of the camera response model;

[0044] Collect a series of images with different exposure ratios taken by the same camera in real scenes, extract the pixel values ​​of the images and fit the function model;

[0045] The generated images with different exposure ratios are compared with the real images with different exposure ratios taken, and the histogram distance between the generated images and the real images is measured. The camera response model is calculated to minimize the histogram distance between the two images.

[0046] In this embodiment, obtaining a generated image for exposure fusion includes: setting the exposure ratio k of the generated image to control the degree of wide dynamic range enhancement, wherein when k=1, the generated image is consistent with the input image; and sending the input image to a camera response model for mapping calculation to obtain a generated image Q corresponding to the exposure ratio k.

[0047] In this embodiment, obtaining the illumination map of the input image includes: converting the input image of size M×N from the RGB color space to the HSV color space, that is, converting the three channels of red, green, and blue into the three channels of hue, saturation, and lightness, and extracting the lightness channel as the lightness map of the image;

[0048] The obtained luminance map of size M×N is downsampled by a factor of s to obtain a luminance map with a resolution of (M / s)×(N / s). A bilateral filter with a radius of r is used to perform edge-preserving smoothing on the luminance map, and then an s-fold upsampling interpolation operation is performed to restore the original image size to obtain the final image illumination map L.

[0049] This embodiment obtains the weight matrix W required for exposure fusion based on the obtained illumination map L. The weight matrix is: W = L μ , where μ controls the degree of enhancement in different brightness intervals;

[0050] The images P′ and Q are exposed and fused, and the weight matrix W is used as the fusion weight coefficient of the pixels of the image P′, and the weight matrix 1-W is used as the fusion weight coefficient of the image Q. The fused image T is obtained, and its fused pixel value calculation formula is as follows:

[0051] T(x,y,c)=W(x,y,c)×P′(x,y,c)+(1-W(x,y,c))×Q(x,y,c),c∈R,G,B

[0052] Among them, x is the horizontal coordinate value of the pixel, y is the vertical coordinate value of the pixel, and c is the corresponding color channel. The image exposure fusion process needs to be applied to the R, G, and B channels. The three channels are weightedly fused separately. The fusion process does not affect each other and is finally merged into an RGB image.

[0053] The fused image is linearly stretched, the maximum value of the RGB three channels of the fused image T is kept unchanged, the minimum value is mapped to the original input image P, and the three-channel linear stretching operation is performed on the middle pixel to obtain the final wide dynamic range processed image Z.

[0054] This embodiment solves the current problem of low regional pixel quality caused by insufficient dynamic range of a single captured image, making the dark details in the captured image more obvious; at the same time, it reduces the computing resource and storage resource consumption of the wide dynamic range processing algorithm, enabling the algorithm to process real-time scenes in a hardware environment.

[0055] Example 2

[0056] At the implementation level, this embodiment provides a specific implementation of a method for processing a wide dynamic range for a single input image. Figure 2 As shown, this embodiment introduces a wide dynamic range processing method for a single input image.

[0057] This implementation first collects a series of images of the same scene with different exposure ratios taken by the same camera, and fits the image pixel values ​​to obtain a camera response model;

[0058] This embodiment takes the beta-gamma camera response model as an example, and the calculation formulas for generating images with different exposure ratios are as follows:

[0059]

[0060] Where g(g) represents the camera response model, β and γ are two parameters of the model, e is the natural coefficient, P is the original image, and images with different exposure ratios are calculated using camera parameters a, b and exposure ratio k.

[0061] This embodiment collects a series of images of the same scene at different exposure ratios taken by the same camera at the same time. Specifically, a static shooting scene is selected, the camera is set to automatic exposure bracketing mode, the exposure compensation amount is manually set to 2EV (2 stops), and more than two images are taken continuously, and the images taken at different exposures are recorded respectively.

[0062] In this embodiment, an exposure ratio k is set, one of the photos is selected, and the values ​​of parameters a and b in the camera response model are continuously adjusted so that the histograms of the generated images with different exposure ratios are closest to the histograms of the corresponding exposure ratio images taken in real life. In this embodiment of the present invention, the distance is measured using the Euclidean distance, and the specific calculation formula is as follows:

[0063]

[0064] Where D(Q,M) represents the histogram distance between the generated image Q and the real image M, 0≤q i ≤1 and 0≤m i ≤1 is the frequency corresponding to the pixel value i. For a 256-level grayscale image, n=256. For an RGB three-channel image, the distance is calculated for each of the three channels, and the final distance is the mean of the three-channel histogram distance.

[0065]

[0066] Among them, D R 、D G 、D B Represent the histogram distances of red, green and blue channels respectively.

[0067] In this embodiment, a series of bracketed exposure images are captured to form a reference image database, and the parameters of the camera response model are fitted and solved. The obtained camera response model minimizes the sum of the histogram distances between the generated image and the images in the database. In this embodiment, the obtained camera response model parameters are a=-0.329 and b=1.126.

[0068] In this embodiment, the exposure ratio k is set, and the input image is used to obtain an additional generated image Q for exposure fusion through the camera response model.

[0069] The exposure ratio controls the degree of wide dynamic range enhancement. When k = 1, the generated image is consistent with the input image. In this embodiment, the exposure ratio k is set to 4. The input image is normalized, and the RGB channels of the image are mapped through the camera response model to obtain the generated image Q.

[0070] This embodiment obtains the brightness map of the original input image P, downsamples it, performs edge-preserving smoothing filtering through bilateral filtering, and then restores the size to obtain the illumination map L corresponding to the input image;

[0071] As an optional implementation, taking a 1920×1080 input RGB three-channel image as an example, the image is converted from the RGB color space to the HSV color space, that is, from the red, green, and blue channels to the hue, saturation, and lightness channels, and the lightness channel is extracted as the image's lightness map;

[0072] In this embodiment, an 8x nearest neighbor interpolation downsampling operation is performed on the obtained luminance map with a size of 1920×1080, that is, a luminance map with a resolution of 240×135 is obtained. A bilateral filter with a radius of 7 is used to perform an edge-preserving smoothing operation on the luminance map. Downsampling the image before performing the edge-preserving smoothing operation can effectively reduce computing resources, and the result is better than the edge-preserving filtering effect without downsampling.

[0073] This embodiment then uses bilinear interpolation to perform 8-fold upsampling to restore the original image size, and obtains the final image illumination map L. The final illumination map can effectively remove the interference of texture information, and objects in the same area have the same illumination.

[0074] In this embodiment, a contrast-limited adaptive histogram equalization operation is performed on the input image, and the contrast enhancement is limited to 0.002, thereby obtaining an image P′ after the contrast of the original image is enhanced.

[0075] In this embodiment, a weight matrix W is obtained based on the illumination map, and the contrast-enhanced image P′ and the generated image Q with different exposure ratios are subjected to exposure fusion. The fused image is linearly stretched to obtain an image Z processed with a wide dynamic range.

[0076] In the embodiment of the present invention, the weight matrix W required for exposure fusion is obtained according to the obtained illumination map L, and the weight matrix is: W = L μ , where μ controls the degree of enhancement in different brightness ranges. In this embodiment, μ=0.5.

[0077] Calculating the weight matrix based on the image's illumination map enables adaptive processing for different exposure conditions in different areas of the image: for poorly exposed areas, the degree of wide dynamic range processing is greater; for areas with lower illumination, the degree of contrast enhancement is less, suppressing dark noise generated during shooting.

[0078] In this embodiment, the image P′ and the image Q are exposed and fused. The weight matrix W is used as the fusion weight coefficient of the image P′ pixel, and the weight matrix 1-W is used as the fusion weight coefficient of the image Q. The fused image T is obtained. The calculation formula of its fused pixel value is as follows:

[0079] T(x,y,c)=W(x,y,c)×P′(x,y,c)+(1-W(x,y,c))×Q(x,y,c),c∈R,G,B

[0080] Among them, x is the horizontal coordinate value of the pixel, y is the vertical coordinate value of the pixel, and c is the corresponding color channel. The image exposure fusion process needs to be applied to the R, G, and B channels. The three channels are weightedly fused separately. The fusion process does not affect each other and is finally merged into an RGB image.

[0081] In this embodiment, the fused image is linearly stretched, the maximum value of the RGB three channels of the fused image T is kept unchanged, the minimum value is mapped to the original input image P, and a three-channel linear stretch operation is performed on the middle pixels to obtain the final wide dynamic range processed image Z.

[0082] In the embodiment of the present invention, a camera response model is constructed to generate images with different exposure ratios for exposure fusion, without the need for images with different exposure ratios shot in real scenes.

[0083] The embodiment of the present invention only requires one low dynamic range image to perform wide dynamic range processing, without the need for multiple shooting operations, thereby reducing the difficulty of shooting and eliminating the need to consider factors such as lens jitter and ghosting caused by shooting images with different exposure ratios.

[0084] The embodiment of the present invention can perform adaptive wide dynamic range processing for different brightness areas, effectively enhance the detail texture of the image, improve the quality of the dark area of ​​the image, and has low computing resources, which can meet the real-time processing requirements of hardware.

[0085] Example 3

[0086] This embodiment provides an image wide dynamic range processing device based on exposure fusion, including:

[0087] The image dataset collection module is used to collect images of the same scene with different exposure ratios taken by the same camera, and store the captured images and their corresponding exposure compensation information;

[0088] The camera response model fitting module is used to construct a mathematical model to express the camera response model, and then fit the collected images with different exposure ratios taken in real scenes;

[0089] An image wide dynamic range processing module includes a memory and a processor, wherein the memory is used to store computer programs, processor input and output data, and related intermediate results. When the processor executes the computer program stored in the memory, the processor performs an image wide dynamic range processing method based on exposure fusion.

[0090] In summary, the present invention generates images with different exposure ratios for exposure fusion by constructing a camera response model, without the need for images with different exposure ratios taken in real scenes. Only one low dynamic range image is needed for wide dynamic range processing, without the need for multiple shooting operations, thus reducing the difficulty of shooting.

[0091] The present invention does not need to consider factors such as lens jitter and ghosting caused by shooting images with different exposure ratios. It can perform adaptive wide dynamic range processing for different brightness areas, effectively enhance the image's detail texture, and improve the quality of the image's dark areas. It also requires low computing resources and can meet the hardware's real-time processing requirements.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for processing an image with a wide dynamic range based on exposure fusion, characterized in that: The method comprises the following steps: S1 collecting multiple images of the same scene shot by the same camera at different exposure ratios, and fitting the image pixel values ​​to obtain a camera response model; S2 inputting the input image into the camera response model to obtain an additional generated image, and obtaining a brightness map of the original input image, and obtaining a lighting map corresponding to the input image after downsampling and processing; obtaining the lighting map of the input image, including: converting the input image with a size of M×N from the RGB color space to the HSV color space, that is, converting the three channels of red, green, and blue into the three channels of hue, saturation, and lightness, extracting the lightness channel as the lightness map of the image, The obtained luminance map of size M×N is downsampled s times to obtain a luminance map with a resolution of (M / s)×(N / s). The luminance map is smoothed edge-preservingly using a bilateral filter with a radius of r, and then an s-fold upsampling interpolation operation is performed to restore the original image size to obtain the final image illumination map L; S3 performs contrast-limited adaptive histogram equalization on the input image to obtain an image with enhanced contrast of the original image; S4 obtains a weight matrix based on the illumination map, and performs exposure fusion on the additional image obtained in S2 and the image of S3, and linearly stretches the image to obtain an image processed with a wide dynamic range.

2. The method for processing wide dynamic range images based on exposure fusion according to claim 1, characterized in that: In the method, a mathematical model is used to construct a camera response model, pixel values ​​of the image are extracted to fit the function model, and the generated images with different exposure ratios are compared with the real images taken with different exposure ratios. The histogram distance between the generated images and the real images is measured, and the camera response model is calculated to minimize the histogram distance between the two images.

3. The image wide dynamic range processing method based on exposure fusion according to claim 2, characterized in that: In the method, a beta-gamma camera response model is selected, and the calculation formula for generating images with different exposure ratios is as follows: ； Where g(g) represents the camera response model, β and γ are two parameters of the model, e is the natural coefficient, P is the original image, and images with different exposure ratios are calculated using camera parameters a, b and exposure ratio k.

4. The method for processing wide dynamic range images based on exposure fusion according to claim 2, characterized in that: Set the exposure ratio k, select one of the photos taken, and continuously adjust the values ​​of parameters a and b in the camera response model so that the histograms of the generated images with different exposure ratios are closest to the histograms of the corresponding exposure ratio images taken in real time. The distance is measured using the Euclidean distance, and the specific calculation formula is as follows: ； Among them, D(Q,M) represents the histogram distance between the generated image Q and the real image M, 0≤qi≤1 and 0≤mi≤1 are the frequencies corresponding to the pixel value i. For RGB three-channel images, the distance is calculated for the three channels respectively, and the final distance is the mean of the three-channel histogram distance ; Among them, DR, DG, and DB represent the histogram distances of the red, green, and blue channels, respectively.

5. The method for processing wide dynamic range images based on exposure fusion according to claim 1, characterized in that: In the method, the exposure ratio of the generated image is set to control the degree of wide dynamic range enhancement, wherein the generated image is consistent with the input image; the input image is sent to the camera response model for mapping calculation to obtain the generated image corresponding to the exposure ratio.

6. The method for processing wide dynamic range images based on exposure fusion according to claim 1, characterized in that: In the method, an input image of size is converted from RGB color space to HSV color space, and the three channels of red, green and blue are converted into three channels of hue, saturation and lightness, and the lightness channel is extracted as the lightness map of the image; a downsampling operation is performed on the obtained lightness map of size to obtain a lightness map of resolution, a bilateral filter with a radius is used to perform an edge-preserving smoothing operation on the lightness map, and then a times upsampling interpolation operation is used to restore the original image size to obtain the final image illumination map.

7. The method for processing wide dynamic range images based on exposure fusion according to claim 1, characterized in that: In the method, the weight matrix W required for exposure fusion is obtained according to the illumination map L. The weight matrix is: W = Lμ, where μ controls the degree of enhancement of different brightness intervals; Image P′ and image Q are exposed and fused, and the weight matrix W is used as the fusion weight coefficient of the image P′ pixel, and the weight matrix 1-W is used as the fusion weight coefficient of the image Q. The fused image T is obtained, and its fused pixel value is calculated as follows: T(x,y,c)=W(x,y,c)×P′(x,y,c)+(1-W(x,y,c))×Q(x,y,c),c∈R,G,B, where x is the pixel horizontal coordinate value, y is the pixel vertical coordinate value, and c is the corresponding color channel. The image exposure fusion process is applied to the R, G, and B channels, and the three channels are weightedly fused separately. The fusion process does not affect each other and is finally merged into an RGB image.

8. The method for processing wide dynamic range images based on exposure fusion according to claim 7, characterized in that: In the method, the fused image is linearly stretched, the maximum value of the RGB three channels of the fused image T is kept unchanged, the minimum value is mapped to the original input image P, and the three-channel linear stretching operation is performed on the middle pixels to obtain the final wide dynamic range processed image Z.

9. An image wide dynamic range processing device based on exposure fusion, characterized in that: include: The image dataset collection module is used to collect images of the same scene with different exposure ratios taken by the same camera, and store the captured images and their corresponding exposure compensation information; A camera response model fitting module is used to construct a mathematical model to express the camera response model, and then fit the collected images with different exposure ratios taken in real scenes; an image wide dynamic range processing module includes a memory and a processor, the memory is used to store computer programs, processor input and output data and related intermediate results, when the processor executes the computer program stored in the memory, the processor executes the image wide dynamic range processing method based on exposure fusion as described in any one of claims 1 to 8.

Images