A method for underwater image enhancement
By combining dark channels, bilateral filtering and Retinex theory, the blur and color casting problems caused by light attenuation and complex water bodies are solved, and the efficient fog removal and color correction of underwater images is achieved, which improves the improvement of the Image clarity and detail retention effect.
Patent Information
- Application Number
- CN202210613693.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-31
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2042-05-31
AI Technical Summary
The existing land image processing methods are not effective underwater underwater, and single dark channel processing cannot effectively solve color imbalance and image imbalance and image imbalance. The overall dark problem.
Using a multi-step approach based on dark channel, bilateral filtering and Retinex theory, including inversion processing, calculating global background light and underwater ambient light, guide filtering processing, multi-scale Retinex algorithm enhancement, color recovery and bilateral filtering noise processing, combined with a histogram Images capture balanced color channels to improve image quality.
It improves the clarity and color authenticity of underwater images, reduces color deviation and noise, effectively realizes details and color recovery, and improves the overall quality of underwater images.
Smart Images

Figure CN115034985B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of image processing, and relates to an underwater image enhancement method, in particular to a method for enhancing the quality of underwater photographed images based on dark channel and Retinex theory. Background Art
[0002] In recent years, with the continuous deepening of ocean exploration at home and abroad, underwater robot technology has been greatly developed, especially unmanned underwater robots, which have gradually become a research hotspot with their low cost and high efficiency. Unmanned underwater robots are usually equipped with visual sensing equipment to capture the surrounding environmental information, with videos, images, live images, etc. as the carriers for recording. However, underwater scenes are very different from land scenes. Even the purest filtered water has serious light attenuation, and the complex water characteristics and a large number of suspended particles will affect the normal transmission of light waves. Images taken underwater usually show a variety of image quality problems such as blur, color cast, and lack of details, which have a serious impact on subsequent image applications.
[0003] At present, the focus of the image processing field is still on land image processing. Although it has some reference significance for underwater image quality enhancement, directly migrating land methods to water will not produce good results. Therefore, it is still necessary to study underwater image quality enhancement methods.
[0004] According to the different image domains considered in the image processing process, underwater image enhancement methods can be roughly divided into two categories: spatial domain methods and transform domain methods. The spatial domain method is based on grayscale mapping and is based on pixel-level image enhancement operations. The more common pixel operations include image contrast enhancement and image grayscale layer improvement. Traditional methods for enhancing image contrast for land images include adaptive histogram equalization, grayscale edge assumption, white balance algorithm, etc. These methods have good effects on land images, but the enhancement effect on underwater images is not ideal. While enhancing the image effect, the spatial domain method will also amplify the problems of color distortion and noise spots. The transform domain method transforms the image from the spatial domain to other domains, such as the frequency domain, wavelet domain, and other feature domains based on some specific transformation rules. The corresponding image enhancement is performed in the feature domain, and then converted back to the spatial domain to obtain the enhanced image. The most common transform domain image processing method is the wavelet transform method.
[0005] Dark channel prior is a common conclusion drawn based on statistics of a large number of clear images - the vast majority of clear and fog-free images have pixels with very low intensity values of at least one color channel in the local area. However, in underwater shooting environments, the light intensity decays rapidly with the increase of water depth, so underwater images are mostly taken in lightless or low-light environments. Underwater images mostly rely on supplementary lighting, and close-range strong lighting also causes large-area reflections in the captured images; in addition, due to the selective absorption of light of different wavelengths by the medium of water, underwater images have serious color cast problems. These problems have led to the poor image enhancement effect of single dark channel processing in underwater scenes, and many land image processing methods are often applied to water, which often results in image color imbalance and overall dark images. Summary of the invention
[0006] In view of the above-mentioned prior art, the technical problem to be solved by the present invention is to provide an underwater image enhancement method based on dark channel, bilateral filtering and Retinex theory, so as to improve the quality of underwater images while taking into account color correction and detail retention.
[0007] In order to solve the above technical problems, the present invention provides an underwater image enhancement method, comprising the following steps:
[0008] Step 1: Invert the original image obtained by underwater photography to obtain the corresponding dark channel image;
[0009] Step 2: Calculate the global background light and underwater ambient light based on the dark channel image;
[0010] Step 3: Calculate the transmittance of the water medium to light after the guided filtering process;
[0011] Step 4: Calculate a preliminary dehazed underwater image based on the global background light and the transmittance of the water medium to light;
[0012] Step 5: Use the multi-scale Retinex algorithm to perform image enhancement on the preliminary dehazed underwater image;
[0013] Step 6: Use color restoration function to compensate for image distortion;
[0014] Step 7: Use the histogram to intercept and balance the proportion of multi-scale Retinex processing values of different color channels in the final processed image to obtain a secondary processed underwater image;
[0015] Step 8: Use the bilateral filter of Gaussian filtering and guided filtering to perform noise removal on the secondary processed underwater image to obtain the final image.
[0016] Furthermore, the inversion process uses a dark channel prior algorithm, including:
[0017] Find the minimum value of each color channel of the original image:
[0018]
[0019] Among them, {R, G, B} represents the three color channels of the image color space, I c (x, y) represents the single color channel image of the c color channel of the original image, c∈{R,G,B}, I dark (x, y) represents the dark channel map of the original image, and Ω represents any local area in the non-pure background area.
[0020] Furthermore, step 2 of calculating the global background light and underwater ambient light according to the dark channel image includes:
[0021] In the dark channel image I dark Select the pixel point whose pixel value is greater than the set threshold in (x, y), determine the pixel value corresponding to the pixel point coordinates in the three RGB channels respectively, and the average of the pixel values corresponding to the pixel point coordinates in each channel is defined as the global background light A ∞ :
[0022]
[0023] Where N is the number of pixels, SUM R , SUM G 、SUM B Respectively represent the sum of the pixel values corresponding to the pixel coordinates in the R, G, and B channels, represents the global background light value on color channel c, c∈{R,G,B};
[0024] Calculate the underwater ambient light for the original underwater image I(x,y):
[0025] A(x,y)=A ∞ [1-e -αd(x,y) ]
[0026] Among them, d(x,y) represents the distance between the pixel point (x,y) and the shooting camera, and α represents the attenuation coefficient of light in water.
[0027] Furthermore, the transmittance of the water medium to light after the guided filtering process in step 3 is:
[0028]
[0029] Among them, k is the set global adjustment parameter, k = α·α λ , α is the attenuation coefficient of light in water, α λis the wavelength coefficient; t(x,y) is the transmittance of light passing through the medium, and the estimated value of t(x,y) satisfies:
[0030]
[0031] Among them, ω∈(0,1) is a set retention coefficient, I c (x,y) represents the single color channel image of the c color channel of the original image. Represents the global background light value on color channel c, c∈{R,G,B}.
[0032] Furthermore, the preliminary dehazing underwater image expression is:
[0033]
[0034] Among them, I(x,y) is the original underwater image, A ∞ is the global background light, is the transmittance of water medium to light after guided filtering, t0 is the set threshold, when When it is less than t0, let GMG k The average grayscale gradient of the image corresponding to the set global adjustment parameter k satisfies:
[0035]
[0036] Among them, P and Q represent the size of the underwater image I(x, y) taken, and the size is expressed as P×Q. The optimal k value is obtained by traversing the loop. The GMG corresponding to the optimal k value is k is the maximum value.
[0037] Furthermore, step 5 uses a multi-scale Retinex algorithm to perform image enhancement processing on the preliminary dehazed underwater image, including:
[0038]
[0039] Among them, R MSRc (x, y) is the value of the MSR algorithm on the c color channel, c∈{R, G, B}; M is the scale value; G m (x, y) represents the surrounding function under different scale constants, m∈M; σ represents the standard deviation of the Gaussian function; ω m represents the weight, and the sum of the weights of the SSR algorithms of M scales should satisfy the normalization condition.
[0040] Furthermore, step 6 uses the color restoration factor to compensate for the image distortion phenomenon as follows:
[0041]
[0042] Among them, R MSRc (x, y) is the value of the MSR algorithm on the c color channel after the image enhancement processing in step 5, c∈{R,G,B}; θ c (x, y) is a color restoration function used to adjust the ratio of the RGB three color channels, which is specifically expressed as:
[0043]
[0044] Among them, β is a constant, which indicates the gain; γ is the parameter for setting the degree of nonlinearity of the control, I c (x, y) represents the single color channel image of the c color channel of the original image, c∈{R, G, B}; then is a single-channel image in color c after color restoration.
[0045] Furthermore, step 7 includes:
[0046] The histogram interception is used, and the expression is:
[0047]
[0048] in, Represents the processing result after weight adjustment. The two constants G and b represent the gain constant and the offset constant respectively. c (x,y) is the color restoration function, I c (x, y) represents the single color channel image of the c color channel of the original image, c∈{R,G,B}, G m (x,y) represents the surrounding function under different scale constants, m∈M.
[0049] Furthermore, step 8 uses a bilateral filter of Gaussian filtering and guided filtering to perform noise processing on the secondary processed underwater image, including:
[0050] The image result H after bilateral filtering using Gaussian filtering and guided filtering is Fin (x,y) is:
[0051]
[0052] Among them, ω(i,j,x,y) represents the weight coefficient; (i,j) represents any point in the window area, traversing the entire area; (x,y) represents the target pixel point, that is, the center point of the area. The value of the weight coefficient ω(i,j,x,y) depends on the product of the definition domain kernel d(i,j,x,y) and the range kernel r(i,j,x,y). The calculation formula of the weight coefficient ω(i,j,x,y) is:
[0053]
[0054] in,
[0055] Beneficial effects of the present invention: The present invention combines the dark color prior theory, which currently performs well in the field of image processing in dehazing processing, with the multi-scale Retinex theory combined with color restoration factors, which has significant effects in the field of color correction, and the bilateral filter, which performs well in the field of noise removal. The invention can improve the quality of underwater images, taking into account the characteristics of color correction, detail retention, and the like. The present invention can have a positive effect on future underwater image processing work, and can serve as a guide for subsequent underwater operations. The processing method of the present invention simultaneously improves the clarity of the original underwater image with foggy color cast and low quality, reduces the degree of color deviation and noise, effectively improves the image quality of the underwater shooting environment, and enhances the details and color restoration of the image. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 It is an overall flow chart of an example of the present invention;
[0057] Figure 2 The figure is a comparison diagram of the image enhancement effect of the present invention, with the original image on the left and the image enhancement effect on the right;
[0058] Figure 3 It is an objective evaluation table of the image enhancement effect of the present invention. DETAILED DESCRIPTION
[0059] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.
[0060] The present invention comprises the following steps:
[0061] Step (1). Input the original underwater image, invert the original image, and obtain the corresponding dark channel image;
[0062] Step (2). According to the relationship between the global background light and the ambient background light, the definition of the global background light is obtained, and the global background light is estimated and calculated;
[0063] Step (3) based on the dark channel image obtained in step (1) and the global background light calculated in step (2), the transmittance of the water medium to light is obtained;
[0064] Step (4). Combine the global background light and light transmittance obtained in step (2) and step (3), and bring them into the restoration formula of the gray gradient value to obtain a preliminary dehazed underwater image;
[0065] Step (5). Perform image enhancement processing on the preliminary dehazed underwater image using a multi-scale Retinex algorithm;
[0066] Step (6). Introduce a color restoration factor and two parameters to compensate for the local distortion caused by the different absorption of light of different wavelengths by water;
[0067] Step (7). Introduce histogram interception to balance the multi-scale Retinex processing values of different color channels and their proportion in the final processed image, control the color cast of the image, and obtain a secondary processed underwater image after color restoration and multi-scale Retinex processing.
[0068] Step (8). Use a bilateral filter of Gaussian filtering and guided filtering to process the underwater image after the secondary processing again.
[0069] Step (9) using objective image evaluation indicators to objectively evaluate the restoration performance of the enhanced underwater images;
[0070] The following is an example with specific parameters:
[0071] Combination Figure 1 , the embodiment of the present invention specifically includes the following steps:
[0072] S1: The original blurred image I (x, y) taken underwater is inverted to obtain a dark channel image corresponding to the original image. The purpose of this step is to eliminate factors in the original image that may affect the light intensity in the shooting environment. In this embodiment, the dark channel prior algorithm is used to invert the original blurred image to obtain a dark channel image. The dark channel prior algorithm is a theory based on statistics of a large number of clear images. Specifically, in the vast majority of clear images, except for the pure background area without a target, in any local area Ω of the non-pure seawater background, there will always be at least one element channel with a very low value, that is, for a clear image, there is a very small minimum value in any local area where a target exists. Based on this principle, the minimum value of each color channel of the original blurred image is obtained.
[0073]
[0074] Among them, {R, G, B} represents the three color channels of the image RGB color space, I c (x, y) represents a single color channel image of a color channel of the RGB color of the original image, I dark (x,y) represents the dark channel image of the original image.
[0075] S2: Estimation calculation of underwater ambient light for the original underwater image I(x,y)
[0076] A(x,y)=A ∞ [1-e -αd(x,y) ] (2)
[0077] Among them, A ∞ represents the global background light underwater, which is an inherent parameter of the water body and a constant; d(x,y) represents the distance between the pixel point (x,y) and the camera; α represents the attenuation coefficient of light in the water body, and the value remains unchanged in the same shooting environment. The brightest points (greater than a certain threshold, or a set value) in the dark channel image are screened out, and the average brightness is calculated to obtain the value of the global background light.
[0078] Select dark channel image I dark The brightest first 0.1% (N) pixels in (x, y) are plotted in the RGB three-channel image I c (x, y), calculate the mean pixel value of the pixel points at these coordinates in the three color channels We define it as the global background light, i.e. Then the value of the underwater ambient light A(x,y) is obtained.
[0079] S3: Dark channel graphics using underwater images I dark (x,y) performs perspective estimation.
[0080] Perspective after guided filtering The expression is:
[0081]
[0082] Among them, t(x,y) is the perspective of light passing through the medium, which can be considered as the water transmittance of the shooting environment; α λ is the wavelength coefficient; k is the set global adjustment parameter, k = α·α λ By referring to the RGB colorimetric system, the wavelengths of red, green and blue light are defined as 700.0nm, 546.1nm and 435.8nm, and α λ =(0.7000,0.5461,0.4358).
[0083] By introducing the dark channel prior theory, the expression for the estimated value of t(x,y) is obtained as follows:
[0084]
[0085] Among them, ω∈(0,1) is a retention coefficient, which is 0.95; represents the global background light value on color channel c. Substituting the value into formula (3) to calculate Numeric value.
[0086] S4.1: Obtain the preliminary processed underwater image J(x,y) after being processed by the dark channel prior theory.
[0087] Based on a classic physical imaging model, the expression of this imaging model is:
[0088] I(x,y)=J(x,y)t(x,y)+A(1-t(x,y)) (5)
[0089] Where J(x,y) represents the fog-free image of the original image I(x,y) after dark channel prior defogging; A represents the shooting environment light, which can usually be considered as the background light of the water environment. The imaging model is transformed to guide the estimated transmittance after filtering Replace t(x,y). Combine S2 and S3 to calculate the global background light A ∞ and the estimated transmittance after guided filtering The image defogging expression can be obtained as:
[0090]
[0091] S4.2: Introduce the image grayscale mean gradient (GMG) as the evaluation standard, and use the traversal loop method to obtain the optimal k value. GMG is used to evaluate the blur degree of an image and is sensitive to the expression of tiny details in the image. The greater the change in grayscale value in a certain direction, the greater the gradient value. The GMG corresponding to a k value k It can be expressed as:
[0092]
[0093] Among them, P and Q represent the size of the underwater image I(x,y) taken, and the size is expressed as P×Q. The restoration formula after introducing the influence of GMG is:
[0094]
[0095] S5: In order to solve the problem that the image processed by the dark channel prior theory is not sufficiently preserved in the relationship between the color channels, the overall image will be dark. In order to enhance the details and color restoration, and preserve the relationship between the color channels, the multi-scale Retinex algorithm is used to remove the fog from the image obtained in S4. To enhance, the formula is as follows:
[0096]
[0097] in, is the value of the MSR algorithm on the c color channel; M is the scale value; Gm (x, y) represents the surrounding function under different scale constants; m∈M; σ represents the standard deviation of the Gaussian function, also known as the scale constant.
[0098] In practical applications, considering the computational efficiency, the value of M is generally between 3 and 6, and 3 is often taken, that is, small scale (σ<50), medium scale (50≤σ<100), large scale (σ≥100); ω m represents the weight, and the sum of the weights of the SSR algorithms of M scales should satisfy the normalization condition.
[0099] S6: Since water absorbs light of different wavelengths differently, images taken underwater often have problems such as color cast and color imbalance. In order to solve the color problem of the image, a color restoration factor is added to compensate for the image distortion. The expression is:
[0100]
[0101] Among them, θ c (x, y) is a color restoration function used to adjust the ratio of the RGB three color channels, which is specifically expressed as:
[0102]
[0103] Among them, β is a constant, indicating the gain; γ is a parameter that controls the degree of nonlinearity. Through multiple experimental comparisons and statistics of most image data, a set of parameter combinations suitable for most underwater images is obtained: γ = 125, β = 46.
[0104] S7: Due to the processing of dark channel prior theory, the proportion of each color channel is inconsistent with the actual situation. In reality, the proportion of the three color channels is not evenly distributed, but the color channels are unevenly distributed according to different images. In order to balance the weights of each color channel, the histogram interception process is introduced, and the expression is:
[0105]
[0106] in, It represents the processing result after weight adjustment, and the two constants G and b represent the gain constant and the offset constant respectively. Through multiple experimental comparisons and statistics of most image data, and referring to the literature data, a set of applicable parameters is obtained: b = -30, G = 192.
[0107] S8: In practical applications, the pixels of underwater cameras are usually not high, the underwater objects are not bright in color, and the edges are easily merged with the background, so the details need to be retained. However, underwater images often contain noise due to suspended particles in the water and ambient light, so it is necessary to use bilateral filtering to process the noise in the captured images.
[0108] The filter combines Gauss filtering and Guided Filtering, with a window size of 7×7, a domain kernel of 20, and a range kernel of 1. The value of the target pixel is calculated as:
[0109]
[0110] Among them, the image result after bilateral filtering is expressed as H Fin (x,y); ω(i,j,x,y) is the weight coefficient; (i,j) represents any point in the window area, and the entire area needs to be traversed; (x,y) is the target pixel, that is, the center point of the area. The value of the weight coefficient ω(i,j,x,y) is determined by the outer product of the domain kernel d(i,j,x,y) and the range kernel r(i,j,x,y). The calculation formula of the weight coefficient is as follows:
[0111]
[0112]
[0113]
[0114] S9: Image after final color processing H Fin (x, y) compared to the original underwater image I(x, y), refer to Figure 2 , the defogging effect is good, the colors are real and rich, the details are clear and noise-free, which has a good impact on subsequent underwater image applications. Objective indicators are evaluated on the underwater images after enhancement, such as Figure 3 As shown. Five objective image quality evaluation indicators are selected: information entropy, grayscale average gradient, peak signal-to-noise ratio, and color difference. Information entropy is the average value of the image grayscale set. When the image grayscale distribution is uniform, the information entropy is the largest. Grayscale average gradient describes the change in grayscale value of the edge of the object in the image. This indicator is mainly used to reflect the clarity and texture coarseness of the image. The larger the value, the clearer the image. Peak signal-to-noise ratio is the most widely used indicator to describe the degree of image distortion. The larger the value, the smaller the image distortion. Color difference is calculated by converting the original color cast image and the corrected image into the LAB color space. The smaller the value, the higher the color restoration.
Claims
1. A method for underwater image enhancement, characterized in that: The following steps are involved: Step 1: Invert the original image obtained by underwater photography to obtain the corresponding dark channel image; Step 2: Calculate the global background light and underwater ambient light based on the dark channel image; Step 3: Calculate the transmittance of the water medium to light after the guided filtering process; Step 4: Calculate a preliminary dehazed underwater image based on the global background light and the transmittance of the water medium to light; Step 5: Use the multi-scale Retinex algorithm to perform image enhancement on the preliminary dehazed underwater image; Step 6: Use color restoration function to compensate for image distortion; Step 7: Use the histogram to intercept and balance the proportion of multi-scale Retinex processing values of different color channels in the final processed image to obtain a secondary processed underwater image; Step 8: Use the bilateral filter of Gaussian filtering and guided filtering to perform noise removal on the secondary processed underwater image to obtain the final image.
2. The underwater image enhancement method according to claim 1, characterized in that: The inversion process adopts a dark channel priori algorithm, including: Find the minimum value of each color channel of the original image: Among them, {R, G, B} represents the three color channels of the image color space, I c (x, y) represents the single color channel image of the c color channel of the original image, c∈{R,G,B}, I dark (x, y) represents the dark channel map of the original image, and Ω represents any local area in the non-pure background area.
3. The underwater image enhancement method according to claim 1, characterized in that: Calculating the global background light and underwater ambient light according to the dark channel image in step 2 includes: In the dark channel image I dark Select the pixel point whose pixel value is greater than the set threshold in (x, y), determine the pixel value corresponding to the pixel point coordinates in the three RGB channels respectively, and the average of the pixel values corresponding to the pixel point coordinates in each channel is defined as the global background light A ∞ : Where N is the number of pixels, SUM R , SUM G , SUM B Respectively represent the sum of the pixel values corresponding to the pixel coordinates in the R, G, and B channels, represents the global background light value on color channel c, c∈{R,G,B}; Calculate the underwater ambient light for the original underwater image I(x,y): A(x,y)=A ∞ [1-e -αd(x,y) ] Among them, d(x,y) represents the distance between the pixel point (x,y) and the shooting camera, and α represents the attenuation coefficient of light in water.
4. The underwater image enhancement method according to claim 1, characterized in that: The transmittance of the water medium to light after the guided filtering process in step 3 is: Among them, k is the set global adjustment parameter, k = α·α λ , α is the attenuation coefficient of light in water, α λ is the wavelength coefficient; t(x,y) is the transmittance of light passing through the medium, and the estimated value of t(x,y) satisfies: Among them, ω∈(0,1) is a set retention coefficient, I c (x,y) represents the single color channel image of the c color channel of the original image. Represents the global background light value on color channel c, c∈{R,G,B}.
5. The underwater image enhancement method according to claim 1, characterized in that: The preliminary defogging underwater image expression is: Among them, I(x,y) is the original underwater image, A ∞ is the global background light, is the transmittance of water medium to light after guided filtering, t0 is the set threshold, when When it is less than t0, let GMG k The average grayscale gradient of the image corresponding to the set global adjustment parameter k satisfies: Among them, P and Q represent the size of the underwater image I(x, y) taken, and the size is expressed as P×Q. The optimal k value is obtained by traversing the loop. The GMG corresponding to the optimal k value is k is the maximum value.
6. The underwater image enhancement method according to claim 1, characterized in that: In step 5, the multi-scale Retinex algorithm is used to perform image enhancement processing on the preliminary defogging underwater image, including: in, is the value of the MSR algorithm on the c color channel, c∈{R, G, B}; M is the scale value; G m (x, y) represents the surrounding function under different scale constants, m∈M; σ represents the standard deviation of the Gaussian function; ω m represents the weight, and the sum of the weights of the SSR algorithms of M scales should satisfy the normalization condition.
7. The underwater image enhancement method according to claim 1, characterized in that: The specific method of using the color restoration function to compensate for image distortion in step 6 is as follows: Among them, R MSRc (x, y) is the value of the MSR algorithm on the c color channel after the image enhancement processing in step 5, c∈{R,G,B}; θ c (x, y) is a color restoration function used to adjust the ratio of the RGB three color channels, which is specifically expressed as: Among them, β is a constant, which indicates the gain; γ is the parameter for setting the degree of nonlinearity of the control, I c (x, y) represents the single color channel image of the c color channel of the original image, c∈{R, G, B}; then is a single-channel image in color c after color restoration.
8. The underwater image enhancement method according to claim 1, characterized in that: Step 7 includes: The histogram is intercepted and the expression is: in, Represents the processing result after weight adjustment. The two constants G and b represent the gain constant and the offset constant respectively. c (x,y) is the color restoration function, I c (x, y) represents the single color channel image of the c color channel of the original image, c∈{R,G,B}, G m (x,y) represents the surrounding function under different scale constants, m∈M.
9. The underwater image enhancement method according to claim 1, characterized in that: The step 8 of performing noise processing on the secondary processed underwater image using a bilateral filter of Gaussian filtering and guided filtering includes: The image result H after bilateral filtering using Gaussian filtering and guided filtering is Fin (x,y) is: Among them, ω(i,j,x,y) represents the weight coefficient; (i,j) represents any point in the window area, traversing the entire area; (x,y) represents the target pixel point, that is, the center point of the area. The value of the weight coefficient ω(i,j,x,y) depends on the product of the definition domain kernel d(i,j,x,y) and the range kernel r(i,j,x,y). The calculation formula of the weight coefficient ω(i,j,x,y) is: in,
Citation Information
Patent Citations
A Multi-Scale Feature Fusion Network based on GANs for Haze Removal
AU2020100274A4
Improved Retinex Image Enhancement Algorithm for Explicit Expression of Multi-scale and Histogram Truncation
AU2020104021A4
Underground pipeline image defogging method based on dark channel and Retinex
CN112862721A
Underwater image enhancement method based on turbidity grading
CN114463211A
