Low-light image enhancement method and mine image enhancement method

Abstract

The invention discloses a low-light-level image enhancement method. The method comprises the following steps: acquiring a target low-light-level image; based on gamma correction, RTV correction and a histogram matching scheme, calculating to obtain a structure diagram and a texture diagram of the target low-light image; according to the obtained structure diagram and texture diagram, based on a Retinex model and a structure / texture guiding scheme, calculating to obtain a brightness diagram and a reflection diagram of the target low-light image; performing adjustment and adaptive correction on the obtained brightness map; and based on the RTV correction scheme, calculating to obtain an enhanced image of the target low-light-level image so as to complete image enhancement of the target low-light-level image. The invention further discloses a mine image enhancement method comprising the low-light-level image enhancement method. According to the method, the enhancement of the low-light image and the mine image is realized, the reliability is higher, the accuracy is better, and the enhancement effect is better.

Description

Technical Field

[0001] This invention belongs to the field of image processing, specifically designing a low-light image enhancement method and a mine image enhancement method. Background Technology

[0002] Low-light images generally refer to images taken in environments with low lighting conditions. Common examples include mine images and surveillance images taken at night. Due to the poor lighting conditions during capture, low-light images generally have poor clarity and resolution. Therefore, image enhancement for low-light images is of great importance.

[0003] Noise and color distortion have always been major challenges in low-light image enhancement. Existing low-light image enhancement schemes are generally based on the Retinex model; these schemes typically perform denoising directly on the reflectance image. However, noise in the reflectance image is often excessively amplified, making the denoising process difficult. Therefore, existing low-light image enhancement schemes often suffer from poor reliability, low accuracy, and poor results.

[0004] Mine images are a typical type of low-light image. Existing image enhancement schemes for mine images also suffer from drawbacks such as poor reliability, low accuracy, and poor results. Summary of the Invention

[0005] One of the objectives of this invention is to provide a low-light image enhancement method that is highly reliable, accurate, and effective.

[0006] The second objective of this invention is to provide a mine image enhancement method that includes the aforementioned low-light image enhancement method.

[0007] The low-light image enhancement method provided by this invention includes the following steps:

[0008] S1. Acquire a low-light image of the target;

[0009] S2. Based on gamma correction, RTV correction and histogram matching scheme, the structure map and texture map of the target low-light image are calculated;

[0010] S3. Based on the structure map and texture map obtained in step S2, calculate the brightness map and reflectance map of the target low-light image using the Retinex model and the structure / texture guidance scheme;

[0011] S4. Adjust and adaptively correct the brightness map obtained in step S3;

[0012] S5. Based on the results obtained in step S4, and using the RTV correction scheme, calculate the enhanced image of the target low-light image to complete the image enhancement of the target low-light image.

[0013] Step S2, which uses a scheme based on gamma correction, RTV correction, and histogram matching to calculate the structure map and texture map of the target low-light image, specifically includes the following steps:

[0014] For the target low-light image obtained in step S1 Preliminary enhancement was performed using gamma correction to obtain a preliminary enhanced image. for ,in This is the gamma correction process;

[0015] For the obtained preliminary enhanced image The image structure was extracted using the RTV correction scheme to obtain the structural map of the target low-light image. for ,in This refers to the RTV correction process.

[0016] For the obtained preliminary enhanced image The histogram matching scheme and CBM3D algorithm were used sequentially to process the image, resulting in a denoised texture map of the target low-light image. for ,in This describes the histogram matching process. This describes the CBM3D algorithm processing procedure.

[0017] Step S3, which involves calculating the brightness and reflectance maps of the target low-light image based on the structure and texture maps obtained in step S2, using the Retinex model and a structure / texture-guided scheme, includes the following steps:

[0018] Based on the structure map and texture map obtained in step S2, the following optimization model is constructed:

[0019] In the formula The brightness map of the target low-light image; The reflectance map of the target in low-light image; This refers to element-wise multiplication of a matrix. Noise map of the target low-light image; For the target low-light image; These are the smoothness weight values; Let Frobenius norm be the matrix. These are the parameters of the structure diagram, and , It is a first-order differential operator. This is a set minimum number to prevent the gradient from being zero; To constrain weight values; For texture map parameters, and ; This is the noise weight value;

[0020] The constructed optimization model is solved to obtain the brightness map and reflectance map of the target low-light image.

[0021] Solving the constructed optimization model includes the following steps:

[0022] For the three independent variables I, R, and N in the optimization model, an alternating optimization scheme is used to solve the problem:

[0023] First, the reflectance map R and the noise map N are treated as constants, and the brightness map I is estimated.

[0024] Then, treating the brightness map I and noise map N as constants, the reflection map R is estimated;

[0025] Finally, the brightness map I and the reflection map R are treated as constants, and the noise map N is estimated.

[0026] Repeat the above estimation process until the set conditions are met; finally, the brightness map and reflectance map of the target low-light image are obtained.

[0027] The solution process includes the following steps:

[0028] (1) Initialization: The value of the initial brightness map is the average value of the R, G, and B channels; the value of the initial reflection map is 1; the value of the initial noise map is 0.

[0029] (2) Setting This represents the brightness map in the k-th iteration. This represents the reflection graph of the k-th iteration. This represents the noise map for the k-th iteration. This represents the initial value of the brightness map calculated in the k-th iteration. This represents the initial value of the reflection map calculated in the k-th iteration;

[0030] (3) Luminance map estimation:

[0031] The optimization equation for the brightness map in the (k+1)th iteration is expressed as:

[0032] In the formula As a weighting factor; Indicates a channel; For target low-light image c-channel data; This represents the c-channel data of the noisy image N in the k-th iteration. This represents the c-channel data of the k-th iteration reflection image R;

[0033] The equation Vectorization: Setting As a matrix-vectorization operator, the brightness map vector i is represented as: The initial value vector of the brightness map calculated in the (k+1)th iteration Represented as Structure diagram parameter vector Represented as ;

[0034] Let D be the Toeplitz matrix obtained from the discrete gradient operator and forward difference, then there exists Define a diagonal matrix for Thus, the vector calculation formula is obtained:

[0035] In the formula Represents the L2 norm;

[0036] Taking the first derivative of the above vector calculation formula and setting it to 0, we get:

[0037] In the formula It is the identity matrix;

[0038] Will Convert back to matrix form to obtain ;

[0039] (4) Estimation of reflectance:

[0040] The optimization equation for the reflection map in the (k+1)th iteration is expressed as:

[0041] In the formula As a weighting factor; The symbol for dividing corresponding elements of a matrix;

[0042] The equation Convert to vector form: Define the reflection map vector Represented as The initial value vector of the reflection map calculated in the (k+1)th iteration Represented as Texture map parameter vector Represented as ;

[0043] Let D be the Toeplitz matrix obtained from the discrete gradient operator and forward difference, then there exists Define a diagonal matrix for Thus, the vector calculation formula is obtained:

[0044] Taking the first derivative of the above vector calculation formula and setting it to 0, we get:

[0045] Will Convert back to matrix form to obtain ;

[0046] (5) Noise map estimation:

[0047] Noise map of the (k+1)th iteration The optimization equation is expressed as:

[0048] In the formula As a weighting factor;

[0049] Direct solution ;

[0050] Repeat steps (3) to (5) above until the set conditions are met: or Or the number of repetitions reaches the set value;

[0051] Finally, the brightness map of the target low-light image is obtained. and reflection diagram .

[0052] Step S4, which involves adjusting and adaptively correcting the brightness map obtained in step S3, specifically includes the following steps:

[0053] Brightness chart adjustment:

[0054] Brightness map After gamma transformation, and then compared with the reflection map Element-wise multiplication of the matrices yields the enhanced brightness map. ;

[0055] Luminance map adaptive correction:

[0056] target low-light image At each pixel, gamma correction is performed to obtain a gamma-corrected luminance map. ;

[0057] When performing gamma correction for each pixel, the adaptively corrected gamma correction value is calculated using the following formula:

[0058] In the formula To obtain a low-light image of the target coordinates Gamma correction value at the location; For target low-light image The average brightness; For target low-light image coordinates The brightness value at that location; The desired target brightness value; This is the global gamma correction value; This is the lower limit of the brightness value; This is the upper limit of the brightness value; This is the adaptively corrected gamma correction value;

[0059] Using adaptively corrected gamma values, the target low-light image is processed in the R, G, and B channels. Perform adaptive brightness correction.

[0060] Step S5, which involves calculating the enhanced image of the target low-light image based on the result obtained in step S4 and the RTV correction scheme, specifically includes the following steps:

[0061] Enhanced brightness map The image structure was extracted using an RTV correction scheme to obtain the first structure map. ;

[0062] Gamma-corrected brightness map The image structure was extracted using an RTV correction scheme to obtain the second structure map. ;

[0063] The enhanced image of the target low-light image is calculated. for .

[0064] The present invention also provides a mine image enhancement method including the aforementioned low-light image enhancement method, specifically including the following steps:

[0065] A. Obtain the mine image to be enhanced;

[0066] B. Using the mine image to be enhanced as the target low-light image, the low-light image enhancement method described above is used for image enhancement;

[0067] C. Obtain the image enhancement result of the mine image to be enhanced, and complete the mine image enhancement.

[0068] The low-light image enhancement method and mine image enhancement method provided by this invention extract the structure map and texture map of the low-light image based on gamma correction, RTV correction and histogram matching scheme. Based on the Retinex model, structure / texture guidance scheme and adaptive correction scheme, it not only enhances the low-light image and mine image, but also has higher reliability, better accuracy and better enhancement effect. Attached Figure Description

[0069] Figure 1 This is a schematic diagram of the process flow for the low-light image enhancement method of the present invention.

[0070] Figure 2 This is a schematic diagram of the process flow of the mine image enhancement method of the present invention.

[0071] Figure 3 This is a schematic diagram illustrating the enhancement effect of the mine image enhancement method of the present invention. Detailed Implementation

[0072] like Figure 1 The diagram shown illustrates the process flow of the low-light image enhancement method of the present invention: This low-light image enhancement method disclosed in the present invention includes the following steps:

[0073] S1. Acquire a low-light image of the target;

[0074] S2. Based on gamma correction, RTV correction, and histogram matching schemes, the structure map and texture map of the target low-light image are calculated; specifically, the following steps are included:

[0075] For the target low-light image obtained in step S1 Because its structure and texture details are not clear enough, gamma correction is first used for preliminary enhancement to obtain a preliminary enhanced image. for ,in This is the gamma correction process;

[0076] For the obtained preliminary enhanced image The image structure was extracted using the RTV correction scheme to obtain the structural map of the target low-light image. for ,in This refers to the RTV correction process.

[0077] For texture extraction, low-light images are inevitably affected by noise, requiring denoising processing. However, the magnitude and type of noise are often unknown. To better denoise, the initial enhanced image is then processed. The histogram matching scheme and CBM3D algorithm were used sequentially to process the image, resulting in a denoised texture map of the target low-light image. for ,in This describes the histogram matching process. This refers to the CBM3D algorithm processing procedure;

[0078] S3. Based on the structure map and texture map obtained in step S2, and using the Retinex model and structure / texture-guided scheme, calculate the brightness map and reflectance map of the target low-light image; including the following steps:

[0079] Texture maps already have good enhancement and denoising effects, but they are usually somewhat smooth; therefore, the method of this invention is based on the Retinex model and the structure / texture guidance scheme for further processing.

[0080] Based on the structure map and texture map obtained in step S2, the following optimization model is constructed:

[0081] In the formula The brightness map of the target low-light image; The reflectance map of the target in low-light image; This refers to element-wise multiplication of a matrix. Noise map of the target low-light image; For the target low-light image; These are the smoothness weight values; Let Frobenius norm be the matrix. These are the parameters of the structure diagram, and , It is a first-order differential operator. This is a set minimum number to prevent the gradient from being zero; To constrain weight values; For texture map parameters, and ; This is the noise weight value;

[0082] In the model, To ensure fidelity, the result of multiplying the decomposed brightness map and reflectance map, plus the noise map, is compared to the original. Figure 1 To; Used to constrain the smoothness of the brightness map; Used to constrain the structure graph, aiming to remove noise while preserving texture; Used to constrain noise levels;

[0083] The constructed optimization model is solved to obtain the brightness map and reflectance map of the target low-light image;

[0084] The solution involves the following steps:

[0085] For the three independent variables I, R, and N in the optimization model, an alternating optimization scheme is used to solve the problem:

[0086] First, the reflectance map R and the noise map N are treated as constants, and the brightness map I is estimated.

[0087] Then, treating the brightness map I and noise map N as constants, the reflection map R is estimated;

[0088] Finally, the brightness map I and the reflection map R are treated as constants, and the noise map N is estimated.

[0089] Repeat the above estimation process until the set conditions are met; finally, the brightness map and reflectance map of the target low-light image are obtained.

[0090] In practice, the following steps can be taken:

[0091] (1) Initialization: The value of the initial brightness map is the average value of the R, G, and B channels; the value of the initial reflection map is 1; the value of the initial noise map is 0.

[0092] (2) Setting This represents the brightness map in the k-th iteration. This represents the reflection graph of the k-th iteration. This represents the noise map for the k-th iteration. This represents the initial value of the brightness map calculated in the k-th iteration. This represents the initial value of the reflection map calculated in the k-th iteration;

[0093] (3) Luminance map estimation:

[0094] The optimization equation for the brightness map in the (k+1)th iteration is expressed as:

[0095] In the formula As a weighting factor; Indicates a channel; For target low-light image c-channel data; This represents the c-channel data of the noisy image N in the k-th iteration. This represents the c-channel data of the k-th iteration reflection image R;

[0096] The equation Vectorization: Setting As a matrix-vectorization operator, the brightness map vector i is represented as: The initial value vector of the brightness map calculated in the (k+1)th iteration Represented as Structure diagram parameter vector Represented as ;

[0097] Let D be the Toeplitz matrix obtained from the discrete gradient operator and forward difference, then there exists Define a diagonal matrix for Thus, the vector calculation formula is obtained:

[0098] In the formula Represents the L2 norm;

[0099] Taking the first derivative of the above vector calculation formula and setting it to 0, we get:

[0100] In the formula It is the identity matrix;

[0101] Will Convert back to matrix form to obtain ;

[0102] (4) Estimation of reflectance:

[0103] The optimization equation for the reflection map in the (k+1)th iteration is expressed as:

[0104] In the formula As a weighting factor; The symbol for dividing corresponding elements of a matrix;

[0105] The equation Convert to vector form: Define the reflection map vector Represented as The initial value vector of the reflection map calculated in the (k+1)th iteration Represented as Texture map parameter vector Represented as ;

[0106] Let D be the Toeplitz matrix obtained from the discrete gradient operator and forward difference, then there exists Define a diagonal matrix for Thus, the vector calculation formula is obtained:

[0107] Taking the first derivative of the above vector calculation formula and setting it to 0, we get:

[0108] Will Convert back to matrix form to obtain ;

[0109] (5) Noise map estimation:

[0110] Noise map of the (k+1)th iteration The optimization equation is expressed as:

[0111] In the formula As a weighting factor;

[0112] Direct solution ;

[0113] Repeat steps (3) to (5) above until the set conditions are met: or Or the number of repetitions reaches a set value; in specific implementation, The preferred value is 0.01, and the preferred setting for the number of repetitions is 20;

[0114] Finally, the brightness map of the target low-light image is obtained. and reflection diagram ;

[0115] S4. Adjust and adaptively correct the brightness map obtained in step S3; specifically including the following steps:

[0116] Brightness chart adjustment:

[0117] Brightness map After gamma transformation, and then compared with the reflection map Element-wise multiplication of the matrices yields the enhanced brightness map. ;

[0118] Luminance map adaptive correction:

[0119] target low-light image At each pixel, gamma correction is performed to obtain a gamma-corrected luminance map. ;

[0120] When performing gamma correction for each pixel, the adaptively corrected gamma correction value is calculated using the following formula:

[0121] In the formula To obtain a low-light image of the target coordinates Gamma correction value at the location; For target low-light image The average brightness; For target low-light image coordinates The brightness value at that location; The desired target brightness value; This is the global gamma correction value; This is the lower limit of the brightness value; This is the upper limit of the brightness value; This is the adaptively corrected gamma correction value;

[0122] Using adaptively corrected gamma values, the target low-light image is processed in the R, G, and B channels. Performing adaptive brightness correction can make the average brightness of the three channels more consistent, thereby improving color distortion and enhancing the image to a suitable brightness.

[0123] During correction, when the pixel value is greater than the image mean, When the value is less than 1, gamma correction is strengthened to increase the pixel value to a higher brightness level than expected; conversely, when the pixel value is less than the image mean, the coefficient... If the value is greater than 1, the enhancement effect weakens, and the enhanced brightness value will be lower than the expected brightness value. Finally, in order to prevent excessive enhancement or excessive weakening, the present invention also introduces upper and lower limits for amplitude limiting.

[0124] S5. Based on the results obtained in step S4, and using the RTV correction scheme, calculate the enhanced image of the target low-light image to complete the image enhancement of the target low-light image; specifically, this includes the following steps:

[0125] Enhanced brightness map The image structure was extracted using an RTV correction scheme to obtain the first structure map. ;

[0126] Gamma-corrected brightness map The image structure was extracted using an RTV correction scheme to obtain the second structure map. ;

[0127] The enhanced image of the target low-light image is calculated. for .

[0128] like Figure 2 The diagram shown is a schematic flowchart of the mine image enhancement method of the present invention: The mine image enhancement method disclosed in this invention, which includes the aforementioned low-light image enhancement method, specifically includes the following steps:

[0129] A. Obtain the mine image to be enhanced;

[0130] B. Using the mine image to be enhanced as the target low-light image, the low-light image enhancement method described above is used for image enhancement;

[0131] C. Obtain the image enhancement result of the mine image to be enhanced, and complete the mine image enhancement.

[0132] like Figure 3 The diagram shows the enhancement effect of the mine image enhancement method of the present invention. Figure 3 As can be seen, the image enhancement method of the present invention can effectively enhance low-light images in a mining environment, and it has high reliability, good accuracy, and good results.

Claims

1. A low-light image enhancement method, comprising the following steps: S1. Obtaining a target low-light image; S2. Calculating a structure map and a texture map of the target low-light image based on a gamma correction, RTV correction and histogram matching scheme; S3. Calculating a brightness map and a reflection map of the target low-light image based on a Retinex model and a structure / texture guide scheme according to the structure map and the texture map obtained in step S2; S4. Adjusting and adaptively correcting the brightness map obtained in step S3; S5. Calculating an enhanced image of the target low-light image based on an RTV correction scheme according to the result obtained in step S4 to complete the image enhancement of the target low-light image.

2. The low-light image enhancement method of claim 1, wherein The structure map and the texture map of the target low-light image calculated based on the gamma correction, the RTV correction and the histogram matching scheme in step S2 specifically comprise the following steps: The target micro-light image acquired in step S1 Preliminary enhancement is performed by using gamma correction to obtain a preliminary enhanced image is wherein is a gamma correction processing process For the obtained preliminary enhanced image , the RTV correction scheme is used to extract the image structure to obtain the structure image of the target glimmer image For , wherein is the RTV correction scheme processing process; The obtained preliminary enhanced image is processed by using a histogram matching scheme and a CBM3D algorithm in sequence to obtain a denoised texture image of the target low-light image is a histogram matching scheme processing process, is a CBM3D algorithm processing process.​​​ 3. The micro- light image intensifying method according to claim 2, wherein The brightness map and the reflection map of the target low-light image calculated based on the Retinex model and the structure / texture guide scheme according to the structure map and the texture map obtained in step S2 in step S3 comprise the following steps: According to the structure map and the texture map obtained in step S2, the following optimization model is constructed: wherein is a luminance map of the target micro light image; is a reflectance map of the target micro light image; is a matrix element multiplication; is a noise map of the target micro light image; is a target micro light image; is a smoothness weight value; is a Frobenius norm of a matrix; is a structure map parameter, and , is a first order differential operator, is a set minimum number to prevent the gradient from being 0; is a constraint weight value; is a texture map parameter, and ; is a noise weight value; The brightness map and the reflection map of the target low-light image are obtained by solving the constructed optimization model.

4. The micro- light image intensifying method according to claim 3, wherein Solving the constructed optimization model comprises the following steps: The three independent variables I, R and N in the optimization model are solved by using an alternating optimization scheme: First, the reflection map R and the noise map N are regarded as constants, and the brightness map I is estimated; Then, the brightness map I and the noise map N are regarded as constants, and the reflection map R is estimated; Finally, the brightness map I and the reflection map R are regarded as constants, and the noise map N is estimated; The above estimation process is repeated until a set condition is reached; Finally, the brightness map and the reflection map of the target low-light image are obtained.

5. The micro- light image intensifying method according to claim 4, characterized in that The solving process specifically comprises the following steps: (1) Initialization: the value of the brightness map is initialized as the average value of the R, G and B channels; the value of the reflection map is initialized as 1; and the value of the noise map is initialized as 0; (2) setting a luminance map of the kth iteration, a reflectance map of the kth iteration, a noise map of the kth iteration, an initial value of the luminance map calculated at the kth iteration, an initial value of the reflectance map calculated at the kth iteration; (3) Brightness map estimation: The optimization equation of the brightness map of the k+1 iteration is represented as: In the formula As a weighting factor; Indicates a channel; For target low-light image c-channel data; This represents the c-channel data of the noisy image N in the k-th iteration. This represents the c-channel data of the k-th iteration reflection image R; The equation Vectorization: Set For the matrix vectorization operator, the intensity map vector i is represented as The initial value vector of the intensity map calculated at the k+1 iteration is represented as The structure map parameter vector is represented as ; Let D be the Toeplitz matrix from the discrete gradient operator and forward difference, then there exists , let the diagonal matrix be ; so the vector calculation formula is obtained: In the formula denotes the L2 norm; The first-order derivative of the above vector calculation formula is calculated, and the first-order derivative is set to 0 to obtain: In the formula is the identity matrix; Convert back to matrix form, we get ; and ; (4) Reflection map estimation: The optimization equation of the reflection map of the k+1 iteration is represented as: wherein is a weight factor; is the sign of the division of the corresponding elements of the matrices. Transforming the equation into vector form: Set the reflectance map vector to be , the initial vector of the reflectance map computed at the k+1 iteration is given by , the texture map parameter vector is given by ; Let D be the Toeplitz matrix from the discrete gradient operator and forward difference, then there exists , let the diagonal matrix be ; so the vector calculation formula is obtained: The first-order derivative of the above vector calculation formula is calculated, and the first-order derivative is set to 0 to obtain: Convert back to matrix form, we get ; and ; (5) Noise map estimation: the noise map of the (k+1)th iteration The optimization equation for the kth iteration is expressed as: In the formula is a weight factor; Directly solving for ; The above steps (3) to (5) are repeated until a set condition is reached: or or the number of repetitions reaches a set value; a luminance map of the target microglint image and a reflection map .

6. The micro- light image intensifying method according to claim 5, wherein The adjustment and adaptive correction of the brightness map obtained in step S3 in step S4 specifically comprise the following steps: Brightness map adjustment: The luminance map is obtained by gamma transformation After gamma transformation, the luminance map is multiplied by the reflection map The matrix element multiplication is performed to obtain the enhanced luminance map ; Brightness map adaptive correction: target micro-light image is obtained At each pixel, gamma correction is performed to obtain a gamma-corrected luminance map ; When performing gamma correction on each pixel, the following formula is used to calculate the gamma correction value after adaptive correction: In the formula is a gamma correction value at a coordinate of a target micro-light image ; is a brightness average value of a target micro-light image ; is a brightness value at a coordinate of a target micro-light image ; is a target brightness value desired to be obtained; is a global gamma correction value; is a lower limit value of a brightness value; is an upper limit value of a brightness value; is an adaptive corrected gamma correction value; Adopting the gamma correction value after self-adaption correction, the target low-light image is carried out brightness self-adaption correction in R, G and B three channels. ​ 7. The micro- light image intensification method according to claim 6, characterized in that The enhanced image of the target low-light image calculated based on the RTV correction scheme according to the result obtained in step S4 in step S5 specifically comprises the following steps: Enhanced brightness map The image structure was extracted using an RTV correction scheme to obtain the first structure map. ; correcting luminance map , extracting image structure using RTV correction scheme to obtain second structure map ; An enhanced image of the target micro-light image is calculated To .

8. A mine image enhancement method including the micro-light image enhancement method according to any one of claims 1 to 7, characterized by Specifically, the following steps are included: A. Obtaining a mine image to be enhanced; B. Taking the mine image to be enhanced as a target low-light image, and using the low-light image enhancement method of any one of claims 1-7 to perform image enhancement. C. obtaining the image enhancement result of the mine image to be enhanced, and completing the enhancement of the mine image.

Images