Robust structure and texture low-light image enhancement algorithm

Abstract

The present invention relates to a robust structure and texture low-light image enhancement algorithm, comprising the following steps: Step 1, reading a low-light image; Step 2, establishing a constraint model for the image; Step 3, setting weights; Step 4, solving the constraint model; Step 5, updating and looping Step 4 until the loop termination condition is met; Step 6, performing gamma correction and solving the enhanced image at the same time. The present invention combines the characteristics of the variational model and the local gradient relative deviation, adopts L2-L P Norms constrain the reflectivity gradient and the illumination component gradient, respectively, and the alternating direction multiplier method is used to simultaneously solve for the object reflectivity and the ambient illumination component. This invention can suppress noise while preserving image detail information, which is of great significance for improving the quality of remote sensing images and enhancing their practical applications in the future.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image processing, and in particular to a robust structure and texture low-illumination image enhancement algorithm. Background Art

[0002] Whether in everyday life or in space remote sensing, the resulting images are affected by multiple factors, including the imaging device and ambient light intensity, leading to image quality degradation. The resulting images exhibit low grayscale and contrast, failing to effectively capture scene details and failing to meet application requirements. Simply increasing image brightness amplifies image noise and reduces the image's dynamic range. Summary of the Invention

[0003] The present invention aims to solve the technical problems in the prior art and provide a robust structure and texture low-light image enhancement algorithm.

[0004] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0005] A robust structure and texture low-light image enhancement algorithm includes the following steps:

[0006] Establish a decomposition constraint model for low-light images:

[0007]

[0008] Among them, R and L represent the object reflectivity and ambient illumination component respectively, S is the low-light image, and N is the noise. and as well as and are the gradients of R and L in the horizontal and vertical directions, respectively, and α, β, and δ are regularization parameters.

[0009] In the above technical solution, the subproblem of R is expressed as:

[0010]

[0011] The solution is:

[0012]

[0013] Among them, l is the vectorization of matrix L, r is the vectorization of matrix R, D lk Let l represent a diagonal matrix:

[0014]

[0015] Among them, G x , G yare represented as Toeplitz matrices of discrete gradient operators with horizontal and vertical positive differences, respectively:

[0016]

[0017] In the above technical solution, the subproblem of L is expressed as:

[0018]

[0019] The solution is:

[0020]

[0021] Among them, l is the vectorization of matrix L, r is the vectorization of matrix R, D rk+1 A diagonal matrix represented by r:

[0022]

[0023] Among them, G x , G y are represented as Toeplitz matrices of discrete gradient operators with horizontal and vertical positive differences, respectively:

[0024]

[0025] In the above technical solution, the sub-problem of N is expressed as:

[0026]

[0027] The solution is:

[0028]

[0029] In the above technical solution, the robust structure and texture low-light image enhancement algorithm further includes the steps of:

[0030] The image constraint model is iterated cyclically to continuously approach the ideal value of optimization until the loop cutoff condition is met, and the object reflectivity R and ambient illumination component L decomposed from the image are output.

[0031] In the above technical solution, the robust structure and texture low-light image enhancement algorithm specifically includes the following steps:

[0032] Step 1, read low-light image;

[0033] Step 2, establish the constraint model of the image;

[0034] Step 3, set weight;

[0035] Step 4, solve the constraint model;

[0036] Step 5, update the loop and execute Step 4 until the loop termination condition is met;

[0037] Step 6: Perform gamma correction to obtain the enhanced image.

[0038] In the above technical solution, in Step 3, the weight settings are specifically as follows: α, β, and δ are set to 0.001-0.005, 0.0001-0.0005, and 0.001 respectively according to needs, γ is set to 2.2, ε is set to 0.001, and γ1 and γ2 are set to 1.5 and 0.05 respectively.

[0039] In the above technical solution, in Step 4, the constraint model to be solved is:

[0040]

[0041] In the above technical solution, in Step 6, gamma correction is The enhanced image is

[0042] The present invention has the following beneficial effects:

[0043] The present invention combines the characteristics of variational model and local gradient relative deviation and adopts L2-L P The norms are used to constrain the reflectivity gradient and the illumination component gradient respectively, and the alternating direction method of multipliers (ADMM) is used to solve the object reflectivity and the ambient illumination component simultaneously.

[0044] The present invention can suppress noise while maintaining image detail information, which is of great significance for improving the quality of remote sensing images and enhancing their practical applications in the later stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Figure 1 The figure is a schematic diagram of the overall flow of the robust structure and texture low-light image enhancement algorithm of the present invention.

[0047] Figure 2 This is a low-light effect image obtained by processing an ordinary natural image using an algorithm in the existing technology.

[0048] Figure 3 This is a diagram showing the enhancement effect of processing ordinary natural images using the robust structure and texture low-light image enhancement algorithm of the present invention.

[0049] Figure 4 This is a low-light effect diagram of remote sensing image processing using algorithms in existing technologies.

[0050] Figure 5 This is a diagram showing the enhancement effect of remote sensing image processing using the robust structure and texture low-light image enhancement algorithm of the present invention. DETAILED DESCRIPTION

[0051] The inventive concept of the present invention is: the present invention proposes a robust structure and texture low-light image enhancement algorithm, which adds corresponding constraints to the low-light image, decomposes the object reflectivity and ambient illumination components, and performs gamma correction on the illumination components to achieve the purpose of image enhancement.

[0052] The present invention will be described in detail below with reference to the accompanying drawings.

[0053] The formula of Retinex principle is expressed as:

[0054]

[0055] Among them, R and L represent the object reflectivity and ambient illumination components respectively, S is the obtained image, Represents element-wise multiplication.

[0056] According to the problem, the present invention adopts a variational model based on the Retinex theory and combines the characteristics of local gradient relative deviation to establish a decomposition constraint model for low-light images:

[0057]

[0058] Among them, R and L represent the object reflectivity and ambient illumination component respectively, S is the low-light image, and N is the noise. and as well as and are the gradients of R and L in the horizontal and vertical directions, respectively, and α, β, and δ are regularization parameters.

[0059] The above problem can be solved by using the Alternating Direction Method of Multipliers (ADMM) to simultaneously solve the object reflectivity and ambient illumination components. The steps are as follows:

[0060] 1. The subproblem of R can be expressed as:

[0061]

[0062] The solution is:

[0063]

[0064] Among them, l is the vectorization of matrix L, r is the vectorization of matrix R, K is the number of iterations; D represents the diagonal matrix, D lk is a diagonal matrix with l after the kth iteration as its entry, D lk The transposed matrix of L and R is updated to the cutoff condition ‖L K -L K-1 ‖ / ‖L K-1 ‖≤ε or ‖R K -R K-1 ‖ / ‖R K-1 Until ‖≤ε is satisfied;

[0065]

[0066] Among them, G x , G y are represented as the Toeplitz matrices of the discrete gradient operators with horizontal and vertical positive differences, respectively.

[0067]

[0068] 2. The subproblem of L can be expressed as:

[0069]

[0070] The solution is:

[0071]

[0072] Among them, l is the vectorization of matrix L, r is the vectorization of matrix R, k is the number of iterations, D rk+1 Let r be a diagonal matrix.

[0073]

[0074] Among them, G x , G y are represented as the Toeplitz matrices of the discrete gradient operators with horizontal and vertical positive differences, respectively.

[0075]

[0076] 3. The subproblem of N can be expressed as:

[0077]

[0078] The solution is:

[0079]

[0080] The image constraint model is iterated repeatedly to continuously approach the ideal value of optimization until the loop cutoff condition (L k -L k-1 ) / (L k-1 )≤εor(R k -R k-1 ) / (R k-1 )≤ε, the object reflectivity R and the ambient illumination component L obtained by decomposing the output image are output. At this time, the illuminance component is gamma corrected, and the expression is as follows:

[0081]

[0082] At the same time, according to the Retinex theory, the enhanced image can be obtained:

[0083]

[0084] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0085] like Figure 1 As shown, the present invention proposes a robust structure and texture low-light image enhancement algorithm, the specific steps of which include:

[0086] Step 1: Read low-light images.

[0087] Step 2: Establish a constraint model for the image.

[0088] Step 3. Set the weights. α, β, and δ are set to 0.001-0.005, 0.0001-0.0005, and 0.001, respectively, according to requirements. γ is set to 2.2, ε is set to 0.001, and γ1 and γ2 are set to 1.5 and 0.05, respectively.

[0089] Step 4, solve the constraint model.

[0090]

[0091] Step 5, update the loop and execute Step 4 until the loop termination condition is met.

[0092] Step 6: Perform gamma correction on the obtained L and obtain the enhanced image.

[0093]

[0094] like Figure 2-5 As shown, the enhanced images obtained by processing ordinary natural images and remote sensing images using the robust structure and texture low-light image enhancement algorithm of the present invention are superior to the low-light processing results of the algorithms in the prior art.

[0095] The present invention combines the characteristics of variational model and local gradient relative deviation and adopts L2-L P The norms are used to constrain the reflectivity gradient and the illumination component gradient respectively, and the alternating direction method of multipliers (ADMM) is used to solve the object reflectivity and the ambient illumination component simultaneously.

[0096] The present invention can suppress noise while maintaining image detail information, which is of great significance for improving the quality of remote sensing images and enhancing their practical applications in the later stages.

[0097] Obviously, the above embodiments are merely examples for clarity of explanation and are not intended to limit the implementation methods. Those skilled in the art will readily appreciate that other variations or modifications based on the above descriptions are possible. It is not necessary and impossible to enumerate all implementation methods here. Obvious variations or modifications arising therefrom remain within the scope of protection of the present invention.

Claims

1. A robust structure and texture low-light image enhancement algorithm, characterized by: The following steps are involved: Establish a decomposition constraint model for low-light images: Among them, R and L represent the object reflectivity and ambient illumination component respectively, S is the low-light image, and N is the noise. and as well as and are the gradients of R and L in the horizontal and vertical directions respectively, α, β and δ are regularization parameters; F and p represent the F norm and p norm respectively, Ω represents the size of the local size of the image, E(R,L) represents a constraint equation about R and L, γ1 and γ2 are exponential parameters about L and R respectively; The decomposition constraint model achieves image enhancement by executing the following steps: Step 1, read low-light image; Step 2, establish the constraint model of the image; Step 3, set weight; Step 4, solve the constraint model; Step 5, update the loop and execute Step 4 until the loop termination condition is met; Step 6: Perform gamma correction to obtain the enhanced image.

2. The robust structure and texture low-light image enhancement algorithm according to claim 1, characterized in that: The subproblem of R is expressed as: The solution is: Among them, l is the vectorization of matrix L, r is the vectorization of matrix R, K is the number of iterations; D represents the diagonal matrix, D lk is a diagonal matrix with l after the kth iteration as its entry, D lk The transposed matrix of L and R is updated to the cutoff condition ‖L K -L K-1 ‖ / ‖L K-1 ‖≤ε or ‖R K -R K-1 ‖ / ‖R K-1 Until ‖≤ε is satisfied; Among them, G x , G y are represented as Toeplitz matrices of discrete gradient operators with horizontal and vertical positive differences, respectively:

3. The robust structure and texture low-light image enhancement algorithm according to claim 1, characterized in that: The subproblem of L is expressed as: The solution is: Among them, l is the vectorization of matrix L, r is the vectorization of matrix R, k is the number of iterations, D rk+1 A diagonal matrix represented by r: Among them, s is the vectorization of matrix S, and n is the vectorization of matrix N; G x , G y are represented as Toeplitz matrices of discrete gradient operators with horizontal and vertical positive differences, respectively:

4. The robust structure and texture low-light image enhancement algorithm according to claim 1, characterized in that: The subproblem of N is expressed as: The solution is:

5. The robust structure and texture low-light image enhancement algorithm according to any one of claims 1 to 4, characterized in that: Also includes the steps: The image constraint model is iterated cyclically to continuously approach the ideal value of optimization until the loop cutoff condition is met, and the object reflectivity R and ambient illumination component L decomposed from the image are output.

6. The robust structure and texture low-light image enhancement algorithm according to claim 1, characterized in that: In Step 3, the weight settings are as follows: α, β, and δ are set to 0.001-0.005, 0.0001-0.0005, and 0.001, respectively, according to requirements; γ is set to 2.2; ε is set to 0.001; and γ1 and γ2 are set to 1.5 and 0.05, respectively.

7. The robust structure and texture low-light image enhancement algorithm according to claim 1, characterized in that: In Step 4, the constraint model to be solved is:

8. The robust structure and texture low-light image enhancement algorithm according to claim 1, characterized in that: In Step 6, gamma correction is The enhanced image is S ′ =R°L ′ .

Images