A highly efficient high dynamic image fusion method

Abstract

The application discloses a kind of high-efficiency high dynamic image fusion methods, belong to single camera subarea high dynamic imaging and time-sharing multi-exposure high dynamic imaging field.First, a group of input image groups are acquired, and rearrangement is carried out, and calibration data is automatically selected from the reordered image group, and a fitting data set is constructed.Then, the coefficients of the camera response function are calibrated according to the fitting data set, and the camera response function is obtained, and based on the camera response function, the scene brightness is reconstructed in the scene irradiation domain using the neighborhood information of the semi-saturation region.Finally, the scene brightness is mapped to the [0,1] interval through tone mapping, and finally the fused image is obtained through the inverse transformation of the camera response function.The application can improve the accuracy of polynomial model calculation and is suitable for most practical application scenarios.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of single camera sub-region high dynamic imaging and sub-time multi-exposure high dynamic imaging, and relates to an efficient high dynamic image fusion method. BACKGROUND

[0002] The multi-exposure image fusion algorithm based on scene reconstruction and tone mapping has less calculation amount and can meet the real-time calculation requirement, and is a commonly used algorithm in current real-time multi-exposure image fusion. The method needs to know the response function of the camera and the exposure ratio of different exposure rate images, and on this basis, high dynamic scene reconstruction and tone mapping are performed to obtain a low-bit high dynamic image for display. The exposure ratio difference between multi-exposure images can be realized by setting different exposure times or setting different light intensity attenuation rates of pixels.

[0003] In terms of camera response function calculation, the commonly used methods are Debevec method and polynomial fitting method, and these two methods have defects in universality and accuracy, respectively. For example, the Debevec method needs accurate camera exposure time and small camera noise, and cannot be applied to cameras with unknown exposure time, and most industrial cameras are usually difficult to accurately calculate the exposure time, so the universality of the method is limited. The polynomial fitting method does not need to introduce the camera exposure time parameter, but maps the scene radiance and the image pixel value into a polynomial function. When this method was first proposed by Mitsunaga and Nayar, the model was sensitive to the initial value of the exposure ratio and had poor convergence. Although Zhang Weixiang et al. improved the polynomial model and set the condition that the pixel value is 0 and the scene brightness is also 0, the model convergence was improved, but the calculation accuracy was not high, mainly in the low pixel value area, the model had high discreteness, and the response function calculation of three channels of color image had large difference.

[0004] The main reason for the dispersion of the current polynomial fitting method in the low pixel value area is that in order to calculate the overall optimization function, only the pixels in the low pixel value area of the low brightness original image can be selected, and the pixels in this area have low signal-to-noise ratio due to the existence of large inherent noise, thereby causing the polynomial model to diverge in this area.

[0005] In the high dynamic scene reconstruction, most of the reconstruction algorithms reconstruct the scene brightness at the same pixel position, and do not consider the spatial relationship of adjacent pixels. Tocci proposed an image reconstruction method based on neighborhood pixels, which arranges the image sequence in descending order of average gray scale, divides the image pixels into well-exposed pixels, semi-saturated region pixels and saturated region pixels according to the threshold value, and then takes the highest exposure image as the reference image. For semi-saturated region and saturated region pixels, the gray value of the corresponding neighborhood in the secondary exposure image is used for reconstruction. However, Tocci method transforms in the image pixel domain, not in the scene brightness domain, and needs to use the camera response function for twice transformation. The camera response function used in Tocci method is calculated based on 12-bit image data, while the bit number of the camera is usually 8, and the maximum value of the pixel domain is 255. The camera response function can only reflect the mapping relationship between the scene brightness and the image pixels within 0-255. After the scene brightness of the secondary exposure image is reconstructed by Tocci method, the scene brightness cannot be mapped to the image pixel domain if it is higher than the maximum scene brightness of the camera response function. Therefore, the ordinary 8-bit camera cannot use Tocci method, but the neighborhood information fusion idea of Tocci method has important reference significance. SUMMARY

[0006] In order to solve the problems of camera response function calculation and image fusion of multi-exposure images under the condition of no accurate exposure time, an efficient high dynamic image fusion method is proposed, the reason why the accuracy of the polynomial fitting method is not high in the low pixel value area is analyzed, and the overall optimization function is improved to local optimization function, so as to ensure that the pixels in the middle area are selected in each image, and the accuracy of the polynomial model calculation is improved.

[0007] The high dynamic image fusion method comprises the following specific steps:

[0008] Step one, obtaining a group of high dynamic images, and rearranging the images in the same position according to the exposure intensity from high to low;

[0009] Specifically,

[0010] Selecting the gray scale of multiple non-saturated image pixels in the same position of the input image group to obtain the average value of the pixel gray scale of each image at the position, rearranging the image group according to the size of the average value, and the calculation formula is as follows:

[0011] g'(I) = Rearr(g(I)) (1)

[0012] g(I) represents the input image group, and g'(I) represents the rearranged image group.

[0013] Step two, automatically selecting calibration data from the rearranged image group to construct a fitting data set;

[0014] Specifically,

[0015] First, the images in the rearranged image group are sequentially grouped two by two, and according to the image pixel flat area gray consistency principle, the center pixel with an absolute error less than 1 pixel in the 4-neighborhood is selected as the fitting data;

[0016] Then, the upper and lower bounds of the pixel gray value are determined, and the fitting data set is constructed in the interval.

[0017] Among them, the lower bound of the pixel gray value is determined according to the signal-to-noise ratio, and the upper bound of the pixel gray value is less than the saturation pixel.

[0018] Step three, according to the fitting data set, the coefficients of the camera response function are calibrated, and the camera response function is obtained;

[0019] The camera response function is:

[0020]

[0021] Where L is the scene brightness, I is the image pixel value, N is the highest order of the polynomial, and N>1, c n is the polynomial coefficient.

[0022] The coefficient calibration of the camera response function is to solve the coefficients N and c n , the specific solving process is as follows:

[0023] Suppose the exposure of the adjacent two images is e q and e q+1 , and the image pixel value is I p,q and I p,q+1 , then

[0024]

[0025] Get the exposure ratio R of the two images:

[0026]

[0027] And R<1.

[0028] Assuming R is known, the difference cost function of the camera response function is constructed as:

[0029]

[0030] Where R q,q+1 represents the exposure ratio of the qth exposure image and the q+1th exposure image.

[0031] By solving the difference cost function, taking the derivative of the coefficient, the coefficients of the camera response function are obtained,

[0032]

[0033] Meanwhile, in order to ensure the convergence of the difference cost function, the camera response function needs to satisfy the following condition:

[0034]

[0035] Through the formula (6) and (7), the linear equation set about the image pixel value and the exposure between images is obtained, and the coefficient of the camera response function is solved.

[0036] The ratio of the average value of the pixels of each adjacent two images in the fitting data set is used as the initial value of iteration The polynomial coefficient is obtained Then the exposure ratio of the kth iteration is updated by the following formula

[0037]

[0038] The stopping iteration condition of the difference cost function is

[0039] ε < δ (9)

[0040] Wherein, δ is a small amount.

[0041] N is obtained when the iteration of the difference cost function stops.

[0042] Step four, based on the camera response function, the scene brightness is reconstructed in the scene irradiance domain using the neighborhood information of the semi-saturation region;

[0043] The specific process is as follows:

[0044] Step 401, selecting a reference image;

[0045] The reference image is not fixed, only the position in the image is fixed, that is, when the selected region in the highest exposure image is saturated and the center pixel of the selected region in the secondary exposure image is saturated but the region is not fully saturated, the secondary exposure image is taken as the reference image, and the neighborhood pixel information thereof is used.

[0046] Step 402, selecting 3x3 pixels as the saturation judgment region of the calculation pixel in the reference image, when the pixel value in the region is greater than the threshold T, it is recorded as a saturated pixel, and the number of saturated pixels N is counted sat According to N sat The type of the calculation pixel in the region is determined, specifically:

[0047] When N sat = 0, the calculation pixel is a well-exposed pixel;

[0048] When 0 < N satWhen N = 9, the computing pixel is determined as a full-saturation region pixel.

[0049] When N sat = 9, the computing pixel is determined as a full-saturation region pixel.

[0050] Step 403, according to the type of the computing pixel, the scene brightness of the selected determination region is calculated in combination with the secondary exposure image;

[0051] ① The exposure is good, and the scene brightness is calculated through the camera response function,

[0052] L p = f(I p,q ) (10)

[0053] ② The saturation region pixel, since there are saturated pixels in the determination region, part of the spatial distribution information in the region is lost, so it is necessary to combine the region and the secondary exposure image for pixel reconstruction.

[0054] When the computing pixel is not saturated, the proportion a of the non-saturated pixels in the determination region is calculated, and the scene brightness calculation formula is,

[0055] L p = a f(I p,q ) + (1-a) f(I p,q+1 ) / R q,q+1 (11)

[0056] When the computing pixel is saturated, the spatial distribution information in the secondary exposure image is used to estimate the center pixel, and then the proportion a of the non-saturated pixels in the determination region is used to calculate.

[0057] The estimated value of the center pixel is,

[0058]

[0059] The scene brightness calculation formula is,

[0060] L p = a f'(I p,q ) + (1-a) f(I p,q+1 ) / R q,q+1 (13)

[0061] ③ The full-saturation region pixel, there is no available spatial information in the determination region. At this time, it is necessary to judge the number N s ' at of saturated pixels in the corresponding region of the secondary exposure image.

[0062] If N s ' at = 0, then the scene brightness calculation formula is,

[0063] L p =f(I p,q+1 ) / R q,q+1 (14)

[0064] If N s ' at If the value is greater than 0, then the corresponding area in the secondary exposure image needs to be used as a reference pixel, and then the scene brightness is reconstructed by combining the pixels in that area in the next exposure image.

[0065] After reconstruction, the expanded scene brightness range is obtained.

[0066] Step 5: Map the scene brightness to the [0,1] range using a global tone mapping algorithm;

[0067] The formula for changing scene brightness is:

[0068] L' p =TMO(L p (15)

[0069] In the formula L' p ∈[0,1] represents the transformed scene brightness, and TMO is the tone mapping operator.

[0070] Step 6: Based on the transformed scene brightness, obtain the fused image I' using the camera response function. p ;

[0071] The calculation formula is

[0072] I' p =f -1 (L' p (16)

[0073] Step 7: For images captured with the same parameters, the fused image is output as a video.

[0074] The advantages and beneficial effects of this invention are as follows:

[0075] 1) This invention designs a practical and reliable high dynamic range imaging method that can handle the camera response function and image fusion problems of multi-exposure image groups without accurate exposure parameters, and is applicable to most practical application scenarios.

[0076] 2) This invention addresses the data selection problem in the camera response function calibration process by designing an automatic selection algorithm, which avoids the problem of selecting only small pixels in low-exposure images and improves the data signal-to-noise ratio and camera response function calibration accuracy.

[0077] 3) The application designs a high dynamic image fusion method, which transforms image data to the scene brightness domain, fully utilizes the spatial information in the semi-saturation field, and makes the saturation boundary transition more natural. BRIEF DESCRIPTION OF DRAWINGS

[0078] Figure 1 It is a schematic diagram of the principle of the high dynamic image fusion method of the application;

[0079] Figure 2 It is a flowchart of the high dynamic image fusion method of the application;

[0080] Figure 3 It is a schematic diagram of scene reconstruction in the embodiment of the application. DETAILED DESCRIPTION

[0081] The application will be further described in detail below with reference to the drawings and embodiments.

[0082] As shown in the drawings, Figure 1 the application adopts a polynomial model to fit the response function of the camera for the scene in which the exposure time of the camera cannot be accurately given. In order to ensure the consistency of the camera response function between the multiple-exposure images, a scene brightness cost function is constructed from multiple multiple-exposure images. In the aspect of data selection, the pixels that need to be calibrated are automatically selected in the specified interval. Meanwhile, in order to avoid the influence of the low pixel value area of the image on the response function due to the reduction of the signal-to-noise ratio, the pixel difference cost function of the scene brightness is calculated by selecting two adjacent images with different exposure rates, so as to ensure that the selected calibration data are located in the middle brightness area. Through iteration of the polynomial coefficients and the exposure ratio between the images, the target function is less than a certain threshold, and the final polynomial of the camera response function and the exposure ratio between the images are obtained. This method can maintain the convergence of the camera response function in the low pixel value area.

[0083] The application is aimed at a group of images arranged from high exposure to low exposure taken by an ordinary 8-bit camera, and is different from the Tocci method which selects one high-exposure reference image and maps the remaining exposure images to the reference image. The method of the application completes scene reconstruction in the radiation domain, and does not determine a unique reference image. The reference image is selected from the area of the secondary exposure according to the saturation condition this time, that is, after the saturation of the first image, the area of the secondary exposure image is selected as the reference image to calculate the fusion weight coefficient.

[0084] An efficient high dynamic image fusion method, as shown in the drawings, Figure 2 specific steps are as follows:

[0085] Step 1: Obtain a group of input images through a high dynamic imaging system, and rearrange the images according to the same position and exposure intensity from high to low;

[0086] Specifically,

[0087] The gray scale of the multiple non-saturated image pixels in the same position in the input image group is selected, the average value of the pixels in the position of each image is obtained, the image group is rearranged according to the size of the average value, and the calculation formula is as follows:

[0088] g'(I) = Rearr(g(I)) (1)

[0089] g(I) represents the input image group, and g'(I) represents the rearranged image group.

[0090] Different from the method of taking the average value of the full image gray scale, the method used in the application avoids the interference of saturated pixels and accelerates the calculation speed.

[0091] Step two, the data selection method of two images is used to automatically select the calibration data from the reordered image group to construct a fitting data set;

[0092] Specifically,

[0093] Firstly, the images in the rearranged image group are grouped two by two in order, and the center pixels with an absolute error less than 1 pixel in the 4-neighborhood are selected as the fitting data according to the gray scale consistency principle of the flat area of the image pixels.

[0094] Then, in order to avoid the low signal-to-noise ratio, a lower bound of the pixel gray scale value is determined, and in order to avoid taking saturated pixels, an upper bound of the pixel gray scale value is determined, and the fitting data set is constructed in the interval.

[0095] Step three, the coefficients of the camera response function are calibrated according to the fitting data set to obtain the camera response function;

[0096] The camera response function has the characteristics of monotonically increasing, and a high-order polynomial is used to fit the camera response function, which is:

[0097]

[0098] Wherein, L is the scene brightness, I is the image pixel value, N is the highest order of the polynomial, and c n is the polynomial coefficient.

[0099] The coefficient calibration of the camera response function is to solve the coefficients N and c n , and the principle is that the camera produces different pixel values for the same scene under different exposures. Assuming that the exposure of the adjacent two images is e q and e q+1 , the image pixel values are I p,q and I p,q+1 , then

[0100]

[0101] The exposure ratio R of the two images is obtained:

[0102]

[0103] Generally, the image exposure needs to be arranged from low to high in the calculation process, so that R < 1.

[0104] In the solving process, it is generally assumed that R is known, and then a difference cost function of the camera response function is constructed:

[0105]

[0106] Where, R q,q+1 represents the exposure ratio of the qth exposure image and the q+1th exposure image.

[0107] By solving the difference cost function, taking the derivative of the coefficient, the coefficient of the camera response function is obtained,

[0108]

[0109] In order to ensure the convergence of the difference cost function, the upper and lower limits of the camera response function need to be limited. In general cases, the camera pixel value of 0 represents no light irradiation, and the pixel value saturation represents the full well of photons. Since the specific number of photons cannot be calibrated when the scene is irradiated to the full well, a relative quantity is used, which is assumed to be 1. Therefore, the coefficients of the camera response function satisfy the following conditions,

[0110]

[0111] Through formulas (6) and (7), a linear equation set about the image pixel value and the exposure between images is obtained, and the polynomial coefficients are solved.

[0112] Since the exposure between images cannot be accurately obtained, the ratio of the average value of the fitting data set can be used as the initial value of iteration The polynomial coefficients are obtained Then the exposure ratio of the kth iteration is updated by the following formula

[0113]

[0114] The stopping iteration condition of the difference cost function is,

[0115] ε < δ (9)

[0116] Where δ is a small quantity. The highest order N is greater than 1.

[0117] Step four, based on the camera response function, the scene brightness is reconstructed in the scene irradiation domain using the neighborhood information of the semi-saturation region;

[0118] The selection of the reference image is as follows: the reference image is not fixed, only the position in the image is fixed, that is, when the selected area in the highest exposure image is saturated and the central pixel in the secondary exposure area is saturated but the area is not fully saturated, the secondary exposure area is taken as the reference image, and the neighborhood pixel information thereof is used.

[0119] In order to improve the operation efficiency of the algorithm, 3*3 pixels of the calculation pixel are selected as the saturation judgment area, and when the pixel value in the area is greater than the threshold T (usually set to 225), the pixel is recorded as a saturated pixel, and the number N of saturated pixels is counted sat . According to N sat , the type of the area is determined, and the calculation pixel is divided into a well-exposed pixel, a semi-saturated area pixel and a fully saturated area pixel. The algorithm calculation steps are as follows:

[0120] ① When N sat =0, the pixel is a well-exposed pixel, and the scene brightness is calculated through the camera response function,

[0121] L p = f(I p,q ) (10)

[0122] ② When 0<N sat <9, the pixel is determined to be a semi-saturated area pixel. Since there are saturated pixels in the judgment area, part of the spatial distribution information in the area is lost, and therefore it is necessary to combine the area and the secondary exposure pixel to reconstruct, corresponding to the reconstruction from area 2 to area 1 in Figure 3

[0123] When the calculation pixel is not saturated, the proportion α of the non-saturated pixels in the judgment area is calculated, and the scene brightness calculation formula is,

[0124] L p =α·f(I p,q )+(1-α)·f(I p,q+1 ) / R q,q+1 (11)

[0125] When the calculation pixel is saturated, the central pixel is first estimated using the spatial distribution information in the secondary exposure, and then the proportion α of the non-saturated pixels in the judgment area is calculated. The estimated value of the central pixel is,

[0126]

[0127] The scene brightness calculation formula is,

[0128] L p =α·f′(I p,q )+(1-α)·f(I p,q+1 ) / R q,q+1 (13)​

[0129] ③ When N sat = 9, the pixel is determined as a full-saturated region pixel, and there is no available spatial information in this region. At this time, it is necessary to determine the number of saturated pixels N s ' at .

[0130] If N s ' at = 0, the scene brightness calculation formula is,

[0131] L p = f(I p,q+1 ) / R q,q+1 (14)

[0132] If N s ' at > 0, it is necessary to take the region of the secondary exposure image as a reference pixel, and then combine the pixels in the region of the next exposure image to reconstruct the scene brightness, corresponding to Figure 3 Region 4 to Region 3 reconstruction, the specific method is the same as step ②. Similarly Figure 3 Region 5 and Region 6 reconstruction method is the same as step ②. After reconstruction, the scene brightness range is greatly expanded, as shown in the envelope of Figure 3 .

[0133] Step five, map the scene brightness to the [0, 1] interval through a global tone mapping algorithm;

[0134] The transformation formula of the scene brightness is:

[0135] L' p = TMO(L p ) (15)

[0136] In the formula, L' p ∈ [0, 1], is the transformed scene brightness, and TMO is a tone mapping operator.

[0137] Step six, based on the transformed scene brightness, obtain the fused image I' p through the camera response function;

[0138] The calculation formula is

[0139] I' p = f -1 (L' p ) (16)

[0140] Step seven, for the images taken with the same parameters, the fused image sequence obtained through the above method can be used as video output.

[0141] The application does not need to provide camera image exposure parameters, is suitable for any arrangement image group photographed by the same camera, automatically selects data and polynomial order by improved polynomial fitting method to fit the camera response function, fully utilizes image neighborhood information to reconstruct the scene brightness of the semi-saturation area, and finally generates high dynamic image through tone mapping. The method has strong adaptability, high algorithm robustness, and richer image reconstruction information, and can be applied to single camera regional high dynamic imaging and time-sharing multi-exposure high dynamic imaging fields.

Claims

1. A highly efficient high dynamic range image fusion method, characterized in that, The specific steps are as follows: Step 1: Acquire a set of high dynamic range images and rearrange the images according to the exposure intensity of the same position from high to low; Step 2: Automatically select calibration data from the rearranged image set to construct the fitting dataset; Specifically: First, the images in the rearranged image group are grouped into pairs in order. Based on the principle of gray-level consistency in flat areas of image pixels, the center pixel with an absolute error of less than 1 pixel in 4 neighborhoods is selected as the fitting data. Then, determine the upper and lower bounds of the pixel grayscale values, and construct a fitting dataset within this range; Step 3: Calibrate the coefficients of the camera response function based on the fitted dataset to obtain the camera response function; The camera response function is: Where L is the scene brightness, I is the image pixel value, N is the highest order of the polynomial, and N > 1, c n These are the polynomial coefficients; The calibration of the camera response function coefficients is equivalent to solving for the coefficients N and c. n The specific solution process is as follows: Assume the exposure of two adjacent images are respectively e q and e q+1 The image pixel values ​​are I p,q and I p,q+1 ,but The ratio of the exposure rates R between the two images is obtained: And R < 1; Based on the exposure ratio R, the difference cost function of the camera response function is constructed as follows: Among them, R q,q+1 This represents the ratio of the exposure rate of the q-th exposure image to the exposure rate of the (q+1)-th exposure image; By solving the difference cost function and taking the derivative of the coefficients, the coefficients of the camera response function can be obtained. Meanwhile, to ensure the convergence of the difference cost function, the camera response function must satisfy the following condition: By combining equations (6) and (7), a system of linear equations concerning image pixel values ​​and inter-image exposure is obtained, from which the coefficients c of the camera response function can be solved. n ; The ratio of the average pixel values ​​of any two adjacent images in the fitted dataset is used as the initial value for iteration. Obtain the polynomial coefficients Then update the exposure ratio. The iteration continues until the difference cost function meets the stopping condition, at which point the coefficient N is obtained; Step 4: Based on the camera response function, reconstruct the scene brightness in the scene irradiance domain using neighborhood information of the half-saturated region; Step 5: Map the scene brightness to the [0,1] range using a global tone mapping algorithm, and obtain the fused image I' based on the transformed scene brightness using the camera response function. p ; The calculation formula is I' p =f -1 (L' p ) (7) L' p The scene brightness after mapping transformation; Step six: For images captured with the same parameters, the fused image is output as a video.

2. The efficient high dynamic range image fusion method according to claim 1, characterized in that, In step one, the method for reordering the images in the input image group is as follows: Select multiple unsaturated pixel gray levels at the same location in the input image group, obtain the average pixel gray level at that location for each image, and rearrange the image group according to the magnitude of the average value. The calculation formula is as follows: g′(I)=Rearr(g(I)) (8) g(I) represents the input image group, and g'(I) represents the rearranged image group.

3. The efficient high dynamic range image fusion method according to claim 1, characterized in that, In step two, the upper and lower bounds of the construction interval of the fitted data set are as follows: the lower bound of the pixel gray value is determined based on the signal-to-noise ratio, and the upper bound of the pixel gray value is less than that of the saturated pixel.

4. The efficient high dynamic range image fusion method according to claim 1, characterized in that, In step three, the updated exposure ratio The formula is: The stopping condition for the difference cost function is as follows: ε<δ (10) Where δ is a small quantity.

5. The efficient high dynamic range image fusion method according to claim 4, characterized in that, In step four, the specific process of reconstructing scene brightness is as follows: Step 401, select a reference image; The reference image is not fixed; only the position in the image is fixed. That is, when the selected area is saturated in the highest exposure image and the selected area center pixel is saturated in the secondary exposure image but the area is not fully saturated, the secondary exposure image is used as the reference image, and its neighboring pixel information is utilized. Step 402: Select a 3×3 pixel area in the reference image as the saturation determination region. When the pixel value in this region is greater than the threshold T, it is recorded as a saturated pixel. Count the number of saturated pixels N. sat According to N sat The type of pixels to be calculated in this region is determined as follows: When N sat When = 0, the calculated pixel is the well-exposed pixel; When 0 <N sat When the value is less than 9, the calculated pixel is determined to be a pixel in a half-saturated region. When N sat When the value is 9, the calculated pixel is determined to be a pixel in the fully saturated region; Step 403: Calculate the scene brightness of the selected judgment area based on the type of the calculated pixels and the secondary exposure image; ① For well-exposed pixels, scene brightness is calculated using the camera response function. L p =f(I p,q ) (11) ② Pixels in the half-saturated region: Since there are saturated pixels in the region, some spatial distribution information in the region is lost. Therefore, it is necessary to combine the region and the secondary exposure image to reconstruct the pixels. When calculating pixel desaturation, the proportion α of unsaturated pixels within the judgment region is calculated, and the scene brightness calculation formula is as follows: L p =α·f(I p,q )+(1-α)·f(I p,q+1 ) / R q,q+1 (12) When calculating pixel saturation, the central pixel is first estimated using the spatial distribution information in the secondary exposure image, and then calculated using the proportion α of unsaturated pixels in the judgment area. The estimated value of the center pixel is, The formula for calculating scene brightness is: L p =α·f′(I p,q )+(1-α)·f(I p,q+1 ) / R q,q+1 (14) ③ If there are no usable spatial information in the fully saturated area, it is necessary to determine the number N′ of saturated pixels in the corresponding area of ​​the secondary exposure image. sat ; If N′ sat =0, then the formula for calculating scene brightness is, L p =f(I p,q+1 ) / R q,q+1 (15) If N′ sat If the value is greater than 0, then the corresponding area in the secondary exposure image needs to be used as a reference pixel, and then the scene brightness is reconstructed by combining the pixels in that area in the next exposure image. After reconstruction, the expanded scene brightness range is obtained.

6. The efficient high dynamic range image fusion method according to claim 5, characterized in that, In step five, the formula for the scene brightness mapping transformation is: L' p < TMO(L p ) (16) In the formula L' p ∈[0,1] represents the transformed scene brightness, and TMO is the tone mapping operator.

Images