A vehicle-mounted infrared image enhancement processing acceleration method and system based on sky detection

Abstract

The present invention discloses a method for accelerating vehicle-mounted infrared image enhancement processing based on sky detection, comprising the following steps: S1, acquiring vehicle-mounted infrared image data; S2, detecting and calibrating the sky region in the infrared image to obtain a sky boundary function; S3, decomposing the image into different frequency components; S4, performing multi-scale saliency extraction operations on the different frequency components to obtain a saliency map; and S5, fusing the saliency extraction results with the original infrared image to obtain an enhanced image E. This method calibrates the sky region, thereby performing data enhancement operations only on non-sky regions. This keeps the pixels in the sky region unchanged, thereby reducing the computation time for a single frame and improving the frame rate of the output image after infrared image enhancement.

Description

Technical Field

[0001] The present invention relates to the technical field of infrared image enhancement, and in particular to a vehicle-mounted infrared image enhancement processing acceleration method and system based on sky detection. Background Art

[0002] When using traditional infrared image processing algorithms to enhance vehicle-mounted infrared images, it is essential to perform contrast enhancement on the ground area to improve the performance of the assisted driving system. However, the focus of vehicle-mounted infrared images is usually on the ground area, in addition to the sky. Enhancing the sky area in infrared images does not improve the performance of the assisted driving system. In vehicle-mounted infrared images, the sky area accounts for a large proportion of the entire image. Traditional infrared image enhancement algorithms require separate calculations for every pixel in the entire image, so they inevitably perform enhancement operations on the sky area. This results in a large waste of computing resources and is not conducive to improving the output image frame rate of the infrared image enhancement module. Summary of the Invention

[0003] The purpose of the present invention is to address the problem that conventional infrared image enhancement algorithms cannot enhance non-sky areas separately when applied to vehicle-mounted infrared images, resulting in a waste of computing resources. A vehicle-mounted infrared image enhancement processing acceleration method based on sky detection is proposed, which not only reduces computing resources but also improves the frame rate of the output image after infrared image enhancement.

[0004] In order to solve the above technical problems, the technical solution of the present invention is:

[0005] A vehicle-mounted infrared image enhancement processing acceleration method based on sky detection includes the following steps:

[0006] S1. Acquire vehicle-mounted infrared image data;

[0007] S2. Detect and calibrate the sky area in the infrared image to obtain the sky boundary function b opt (x);

[0008] S3, decompose the image into different frequency components d1, d2, ..., d m+1 ;

[0009] S4. Perform multi-scale saliency extraction on all frequency components to obtain a saliency map.

[0010] S5. Fusing the saliency extraction result with the original infrared image to obtain the enhanced image E.

[0011] Preferably, step S2 includes the following sub-steps:

[0012] S2-1, converting the infrared image into a grayscale image;

[0013] S2-2, take the grayscale image as input and convolve each pixel in the image through the Sobel operator to obtain the gradient map grad of the entire image;

[0014] S2-3. Input the gradient image grad and obtain the boundary position function b of the sky area through the preliminary calculation method of the sky boundary. opt (x), where x represents the horizontal coordinate of the pixel point in the entire image.

[0015] Preferably, in step S2-3, the sky boundary is preliminarily calculated as follows:

[0016] Calculate the threshold t once, the formula is:

[0017]

[0018] The values ​​of n, thresh_min, and thresh_max will affect the final effect. In actual applications, it is necessary to select the best values ​​according to different application scenarios.

[0019] For the currently calculated t and gradient map grad, let the variable x loop from 1 to W: initialize the array b tmp All elements of (x) are H. For each x, do another cycle and let y go from 1 to H: if grad(x, y) > t, then b tmp (x) is set to y and ends the current y loop, and enters the next x loop until all x and y loops are completed, where x represents the horizontal coordinate of the pixel point in the entire image, y represents the vertical coordinate of the pixel point in the image, W is the horizontal width of the image, and H is the vertical height of the image;

[0020] According to b tmp (x) Calculate the energy function J of the image n , the formula is:

[0021]

[0022] Among them, ∑ s and ∑ G are the covariance matrices of the grayscale values ​​of each pixel belonging to the sky area and the ground area, respectively. Both are calculated by the 3x3 matrix centered at the current pixel point. s and λ G are the eigenvalues ​​of the two covariance matrices mentioned above;

[0023] If J n >J n_max , then update Jn_max The value of J n Assign the value of J n_max , and b opt (x) is updated to b at this time tmp (x);

[0024] For the parameter n, let it start from 1 and accumulate 1 each time. When the loop of n is completed, output b opt (x).

[0025] Preferably, the step S2 further includes step S2-4:

[0026] False sky area correction, input is b opt (x), the output is the corrected b opt (x);

[0027] Calculate border_ave using the following formula:

[0028]

[0029] To calculate ASADSBP, the formula is as follows:

[0030]

[0031] Determine the result of the following formula:

[0032] (border_ave <thresh1)&(borde_ave<thresh2|ASADSBP> thresh3)

[0033] Where thresh1=H / 30, thresh2=H / 10, thresh3=5.

[0034] If the output of the above formula is 1, then this sky area is a false sky area, let b opt (x)=H, if the output of the above formula is 0, then b opt (x) remains unchanged.

[0035] Preferably, in step S3, the graph decomposition formula is as follows:

[0036]

[0037] Where fk is the image obtained by smoothing the image using the kth λ, λ represents the smoothing factor, f0 is the input image O before smoothing, d m+1 f m , d k =f k-1 -f k, use m smoothing factors to smooth the image, and after decomposing the image, we can get m+1 layers of images, and define the kth layer image as d k .

[0038] Preferably, the image smoothing operation adopts an image smoothing operation method based on L0 norm minimization.

[0039] Preferably, step S4 includes the following sub-steps:

[0040] According to b opt (x) Determine d1, d2, ..., d m+1 Whether each pixel in the m+1 image components belongs to the sky area, when judging the pixel (x0, y0), if y0>b opt (x0) then the pixel belongs to the sky area, if y0≤b opt (x0), the pixel belongs to the non-sky area;

[0041] Select N windows, the side length of the basic window is t0, the step length of the side length between windows is Δt, and the side length of the nth (1≤n≤N) window is t=t0+nΔt;

[0042] For each component of the image, the pixels belonging to the non-sky area are subjected to the frequency-tuned saliency extraction operation using the N windows mentioned above as units, and the pixels belonging to the sky area are replaced with their own grayscale values; for the kth component d k After performing saliency extraction on the nth window, the saliency map can be obtained. The extraction process can be expressed as:

[0043]

[0044] Finally, (m+1)N saliency maps can be obtained

[0045] Preferably, step S5 includes the following sub-steps:

[0046] S5-1. First, multiply each saliency map with its corresponding image component to obtain the corresponding intermediate quantity

[0047] S5-2. Each image component has N intermediate quantities, and the largest one is selected. When the kth component d k When calculating, the formula is as follows:

[0048]

[0049] S5-3. Obtain the kth image component d by the following formula k The corresponding enhanced image:

[0050]

[0051] Where w1 and w2 are coefficients between 0 and 1. The original image has m+1 image components, so a total of E 1 , E 2 ,...,E m+1 A total of m+1 enhanced images;

[0052] S5-4, then E 1 , E 2 ,...,E m+1 and a set of coefficients α1, α2, ..., α between 0 and 1 m+1 After multiplication, the enhanced image of the whole image is combined, which is expressed by the following formula:

[0053] E=α1E 1 +α2E 2 +...+α m+1 E m+1

[0054] E is the enhanced image of the entire image.

[0055] The present invention also provides a system for implementing the above-mentioned vehicle-mounted infrared image enhancement processing acceleration method based on sky detection, comprising a sky detection preprocessing module, an image decomposition module, a saliency extraction module, and an image enhancement module.

[0056] The sky detection preprocessing module is used for sky area detection and calibration to obtain the sky boundary function b opt (x);

[0057] The image decomposition module is used to decompose the image into different frequency components d1, d2, ..., d m+1 ;

[0058] The saliency extraction module performs multi-scale local saliency extraction operations on the different frequency components obtained by the image decomposition module to obtain a saliency map.

[0059] The image enhancement module is used to improve the contrast of the image and fuse the obtained saliency map with the original image to obtain an enhanced image.

[0060] The present invention has the following characteristics and beneficial effects:

[0061] This technical solution calibrates the sky area by preprocessing the data with sky detection before it enters the infrared image enhancement module. Once the data enters the infrared image enhancement module, only the calibrated pixels in the non-sky area are enhanced, leaving the pixels in the sky area unchanged. This reduces the processing time for a single frame and improves the frame rate of the enhanced infrared image output. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] 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.

[0063] Figure 1 The present invention provides a flowchart of a vehicle-mounted infrared image enhancement processing acceleration method based on sky detection according to an embodiment of the present invention.

[0064] Figure 2 This is a structural block diagram of an embodiment of the present invention.

[0065] Figure 3 is the current frame of the input image captured by the vehicle-mounted infrared camera.

[0066] Figure 4 Schematic diagram for detecting and calibrating the sky area in infrared images. DETAILED DESCRIPTION

[0067] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0068] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings.

[0069] On the contrary, the present invention covers any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention as defined by the claims. Furthermore, to facilitate a better understanding of the present invention, certain specific details are described in detail below in the detailed description of the present invention. Those skilled in the art will be able to fully understand the present invention without these details.

[0070] Example 1

[0071] This embodiment provides a vehicle-mounted infrared image enhancement processing acceleration method based on sky detection, such as Figure 1 As shown, the following steps are included:

[0072] S1. Obtain vehicle-mounted infrared image data.

[0073] It is understandable that the image used for data enhancement is collected by the vehicle-mounted infrared camera. Figure 3 The current frame of the input image captured by the vehicle-mounted infrared camera.

[0074] S2. Detect and calibrate the sky area in the infrared image to obtain the sky boundary function b opt (x), in this embodiment, after the following steps, the function can be obtained as Figure 4 Indicated by the red line.

[0075] Specifically, it includes the following sub-steps:

[0076] S2-1, the input image is a grayscale image obtained from an infrared camera, and the calculation process is to convolve each pixel in the image through the Sobel operator to obtain the gradient map grad of the entire image;

[0077] It should be noted that obtaining the gradient map grad by using the Sobel operator is a conventional technical means in image processing technology, and therefore will not be described in detail in this embodiment.

[0078] S2-2. Construct a preliminary calculation module for the sky boundary. The input is the gradient map grad obtained in S2-1, and the output is the boundary position function b of the sky area. opt (x), x represents the horizontal coordinate of the pixel point in the entire image, and y represents the vertical coordinate of the pixel point in the image.

[0079] Specifically, the boundary position function b of the sky area opt The method for obtaining (x) comprises the following steps:

[0080] (1) Calculate the threshold t once, the formula is:

[0081]

[0082] It should be noted that the values ​​of n, thresh_min, and thresh_max will affect the final effect. In actual applications, it is necessary to select the best values ​​according to different application scenarios.

[0083] In this embodiment, the parameters n=120, thresh_min=5, thresh_max=600, J n_max =0, W is the horizontal width of the image, H is the vertical height of the image;

[0084] (2) For the currently calculated t and gradient map grad, let the variable x loop from 1 to W: initialize the array btmp (x) is H. For each x, do another loop and let y go from 1 to H: if grad(x, y) > t, then b tmp (x) is set to y and ends the current y loop, then enters the next x loop until all x and y loops are completed;

[0085] (3) According to b tmp (x) Calculate the energy function J of the image n , the formula is:

[0086]

[0087] Among them, ∑ s and ∑ G are the covariance matrices of the grayscale values ​​of each pixel belonging to the sky area and the ground area, respectively. Both are calculated by the 3x3 matrix centered at the current pixel point. s and λ G are the eigenvalues ​​of the two covariance matrices mentioned above;

[0088] (4) If J n >J n_max , then update J n_max The value of J n Assign the value of J n_max , and b opt (x) is updated to b at this time tmp (x); output b when the loop of n is finished opt (x).

[0089] When looping the above steps again, for the parameter n, start it from 1 and add 1 each time until it reaches 120 and stops.

[0090] S2-3, false sky area correction, input is b opt (x), the output is the corrected b opt (x);

[0091] Calculate border_ave using the following formula:

[0092]

[0093] To calculate ASADSBP, the formula is as follows:

[0094]

[0095] Determine the result of the following formula:

[0096] (border_ave<thresh1)&(borde_ave<thresh2|ASADSBP>thresh3)

[0097] In this embodiment, thresh1=H / 30, thresh2=H / 10, and thresh3=5.

[0098] If the output of the above formula is 1, then this sky area is a false sky area, let b opt (x)=H. If the output of the above formula is 0, then b opt (x) remains unchanged.

[0099] It can be understood that in the above technical solution, by further performing false sky area correction, the accuracy of sky area calibration is further improved, thereby ensuring the quality of the enhanced image.

[0100] S3, decompose the image into different frequency components d1, d2, ..., d m+1 .

[0101] Specifically, the input is the original image, and the output is the decomposed image d1, d2, ..., d m+1 , the original image is operated on by an image smoothing method based on minimizing the L0 norm.

[0102] It should be noted that the image smoothing method based on L0 norm minimization is a general image smoothing method. After smoothing the image, the formula can be obtained:

[0103] f=L(O,λ)

[0104] f represents the smoothed image, O represents the original image, and λ represents the smoothing factor. As the value of λ increases, the smoothing effect will be stronger.

[0105] In this module, the image is smoothed using m gradually increasing λ values ​​to obtain m smoothed images f1, f2, ..., f m , when the kth λ is used for calculation, it can be expressed as:

[0106] f k =L(O,λ k )

[0107] Subtract the smoothed images one by one to decompose the image to obtain m-1 layers of images, and define the kth layer image as d k , calculated by the following formula:

[0108] d k =f k-1 -f k

[0109] At the same time, let f0=O, d m+1 =f m , d m+1 with d1, ..., d m After merging, we get m+1 components d1, d2, ..., d m+1 , which can be expressed as:

[0110]

[0111] S4. Perform multi-scale saliency extraction on all frequency components to obtain a saliency map.

[0112] Specifically, according to b opt (x) Determine whether each pixel belongs to the sky area. When judging the pixel (x0, y0), if y0>b opt (x0) then the pixel belongs to the sky area, if y0≤b opt (x0) then the pixel belongs to the non-sky area. In this embodiment, Figure 4 The red line in the middle is the sky boundary area function b opt (x), so the area above the red line belongs to the sky area, and the area below the red line belongs to the non-sky area.

[0113] Select N windows, the side length of the basic window is t0, the step length of the side length between windows is Δt, and the side length of the nth (1≤n≤N) window is t=t0+nΔt;

[0114] For each component of the image, the pixels belonging to the non-sky area are subjected to the saliency extraction operation based on frequency tuning using the N windows mentioned above. This is a general saliency extraction method. The pixels belonging to the sky area are replaced by their own grayscale values. For the kth component d k After performing saliency extraction on the nth window, the saliency map can be obtained. The extraction process can be expressed as

[0115]

[0116] Finally, (m+1)N saliency maps can be obtained

[0117] Since only the pixels in the non-sky area will be subjected to the above frequency tuning-based saliency extraction operation, the pixels in the sky area will not be subjected to the above frequency tuning-based saliency extraction operation and will be replaced by their own grayscale values. Figure 4The pixels below the red line in the middle are processed with frequency-tuned saliency extraction. When using traditional image enhancement methods, the entire image is processed. Therefore, in this embodiment, a large amount of computing resources can be saved and the saliency map can be completed in less time. The output of the image can be enhanced to improve the output image frame rate of the entire image enhancement process.

[0118] S5. Fusion of saliency extraction results with original infrared images

[0119] The enhanced image E is obtained.

[0120] Specifically, the (m+1)N saliency maps obtained by input The output is the enhanced image;

[0121] First, multiply each saliency map with its corresponding image component to obtain the corresponding intermediate quantity When the kth component d k Perform saliency map of the nth window When calculating , the calculation formula is as follows:

[0122]

[0123] Each image component corresponds to N intermediate quantities, from which the largest one is selected. When the kth component d k When calculating, the formula is as follows:

[0124]

[0125] Then the kth image component d is obtained by the following formula k The corresponding enhanced image:

[0126]

[0127] Where w1 and w2 are coefficients between 0 and 1. The original image has m+1 image components, so a total of E 1 , E 2 ,...,E m+1 There are m+1 enhanced images in total, and then E 1 , E 2 ,...,E m+1 and a set of coefficients α1, α2, ..., α between 0 and 1 m+1 After multiplication, the enhanced image of the whole image is combined, which is expressed by the following formula:

[0128] E=α1E 1 +α2E 2 +...+α m+1 E m+1

[0129] E is the enhanced image of the entire image.

[0130] Example 2

[0131] This embodiment provides a system for implementing the vehicle-mounted infrared image enhancement processing acceleration method based on sky detection described in Example 1, including a sky detection preprocessing module, an image decomposition module, a saliency extraction module, and an image enhancement module.

[0132] The sky detection preprocessing module is used to perform step S2 in embodiment 1 to detect and calibrate the sky area to obtain the sky boundary function b opt (x);

[0133] The image decomposition module is used to perform step S3 in embodiment 1 to decompose the image, and decompose the image into different frequency components d1, d2, ..., d m+1 ;

[0134] The saliency extraction module is used to perform step S4 in embodiment 1 to perform multi-scale local saliency extraction operations on the different frequency components obtained by the image decomposition module to obtain a saliency map.

[0135] The image enhancement module is used to execute step S5 in embodiment 1 to improve the contrast of the image, and fuse the obtained saliency map with the original image to obtain an enhanced image.

[0136] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions, and variations of these embodiments, including components, without departing from the principles and spirit of the present invention are still within the scope of protection of the present invention.

Claims

1. A vehicle-mounted infrared image enhancement processing acceleration method based on sky detection, characterized in that: The steps include: S1. Acquire vehicle-mounted infrared image data; S2. Detect and calibrate the sky area in the infrared image to obtain the sky boundary function b opt (x); S3, decompose the image into different frequency components d1, d2, ..., d m+1 ; S4. Perform multi-scale saliency extraction on all frequency components to obtain a saliency map. The step S4 includes the following sub-steps: According to b opt (x) Determine d1, d2, ..., d m+1 Whether each pixel in the m+1 image components belongs to the sky area, when judging the pixel (x0, y0), if y0>b opt (x0) then the pixel belongs to the sky area, if y0≤b opt (x0), the pixel belongs to the non-sky area; Select N windows, the side length of the basic window is t0, the step length of the side length between windows is Δt, and the side length of the nth (1≤n≤N) window is t=t0+nΔt; For each component of the image, the pixels belonging to the non-sky area are subjected to the frequency-tuned saliency extraction operation using the N windows mentioned above as units, and the pixels belonging to the sky area are replaced with their own grayscale values; for the kth component d k After performing saliency extraction on the nth window, the saliency map can be obtained. The extraction process can be expressed as: Finally, (m+1)N saliency maps can be obtained S5. Fusing the saliency extraction result with the original infrared image to obtain the enhanced image E.

2. The vehicle-mounted infrared image enhancement processing acceleration method based on sky detection according to claim 1 is characterized in that: The step S2 includes the following sub-steps: S2-1, converting the infrared image into a grayscale image; S2-2, take the grayscale image as input, and convolve each pixel in the image through the Sobel operator to obtain the gradient map grad of the entire image; S2-3. Input the gradient image grad and obtain the boundary position function b of the sky area through the preliminary calculation method of the sky boundary. opt (x), where x represents the horizontal coordinate of the pixel point in the entire image.

3. The vehicle-mounted infrared image enhancement processing acceleration method based on sky detection according to claim 2 is characterized in that: In step S2-3, the sky boundary is initially calculated as follows: Calculate the threshold t once, the formula is: For the currently calculated t and gradient map grad, let the variable x loop from 1 to W: initialize the array b tmp For each element of x, we make another loop and let y go from 1 to H: if grad(x, y) > t, then b tmp (x) is set to y and ends the current y loop, and enters the next x loop until all x and y loops are completed, where x represents the horizontal coordinate of the pixel point in the entire image, y represents the vertical coordinate of the pixel point in the image, W is the horizontal width of the image, and H is the vertical height of the image; According to b tmp (x) Calculate the energy function J of the image n , the formula is: Among them, ∑ s and ∑ G are the covariance matrices of the grayscale values ​​of each pixel belonging to the sky area and the ground area, respectively. Both are calculated by the 3x3 matrix centered at the current pixel point. s and λ G ∑ s and ∑ G The characteristic value of If J n >J n_max , then update J n_max The value of J n Assign the value of J n_max , and b opt (x) is updated to b at this time tmp (x); For the parameter n, let it start from 1 and accumulate 1 each time. When the loop of n is completed, output b opt (x).

4. The vehicle-mounted infrared image enhancement processing acceleration method based on sky detection according to claim 3 is characterized in that: The step S2 further includes step S2-4: False sky area correction, input is b opt (x), the output is the corrected b opt (x); Calculate border_ave using the following formula: Among them, border_ave refers to the average position of the sky boundary; To calculate ASADSBP, the formula is as follows: Among them, ASADSBP refers to the average value of the sum of absolute differences of the sky boundary positions; Determine the result of the following formula: (border_ave<thresh1)&(borde_ave<thresh2|ASADSBP>thresh3) If the output of the above formula is 1, then this sky area is a false sky area, let b opt (x)=H, if the output of the above formula is 0, then b opt (x) remains unchanged, Among them, thresh1, thresh2 and thresh3 are three key thresholds used to determine the sky boundary. These three thresholds need to be pre-set to specific values ​​based on the usage effects of specific application scenarios; W is the horizontal width of the image; H is the vertical height of the image.

5. The vehicle-mounted infrared image enhancement processing acceleration method based on sky detection according to claim 4 is characterized in that: In step S3, the image decomposition formula is as follows: Among them, f k is the image obtained after image smoothing using the kth λ, λ represents the smoothing factor, f0 is the input image O before smoothing, d m+1 f m , d k =f k-1 -f k , use m smoothing factors to smooth the image, and after decomposing the image, we can get m+1 layers of images, and define the kth layer image as d k , Finally, we can get d1, d2, ..., d m+1 Layer images, which are the decomposed images.

6. The vehicle-mounted infrared image enhancement processing acceleration method based on sky detection according to claim 5 is characterized in that: The image smoothing operation adopts an image smoothing operation method based on L0 norm minimization.

7. The vehicle-mounted infrared image enhancement processing acceleration method based on sky detection according to claim 5 is characterized in that: The step S5 includes the following sub-steps: S5-1. First, multiply each saliency map with its corresponding image component to obtain the corresponding intermediate quantity S5-2. Each image component has N intermediate quantities, and the largest one is selected. When the kth component d k When calculating, the formula is as follows: S5-3. Obtain the kth image component d by the following formula k The corresponding enhanced image: Where w1 and w2 are coefficients between 0 and 1. The original image has m+1 image components, so a total of E 1 , E 2 ,...,E m+1 A total of m+1 enhanced images; S5-4, then E 1 ,E 2 ,...,E m+1 and a set of coefficients α1,α2,...,α between 0 and 1 m+1 After multiplication, the enhanced image of the whole image is combined, which is expressed by the following formula: E=α1E 1 +α2E 2 +…+a m+1 E m+1 E is the enhanced image of the entire image.

8. A system for implementing the vehicle-mounted infrared image enhancement processing acceleration method based on sky detection according to any one of claims 1 to 7, characterized in that: It includes vehicle-mounted infrared image acquisition module, sky detection preprocessing module, image decomposition module, saliency extraction module, and image enhancement module. The vehicle-mounted infrared image acquisition module is used to acquire vehicle-mounted infrared images to obtain original images; The sky detection preprocessing module is used for sky area detection and calibration to obtain the sky boundary function b opt (x); The image decomposition module is used to decompose the image into different frequency components d1, d2, ..., d by image smoothing based on L0 norm minimization. m+1 ; The saliency extraction module performs multi-scale local saliency extraction operations on the different frequency components obtained by the image decomposition module to obtain a saliency map. The image enhancement module is used to improve the contrast of the image and fuse the obtained saliency map with the original image to obtain an enhanced image.

Images