Low-light image enhancement method, system, terminal and storage medium

Abstract

The application relates to a low-light image enhancement method, system, terminal and storage medium. The method comprises the following steps: establishing a new visual imaging model according to a traditional Retinex visual imaging model; estimating an initial value of an illumination component according to the established new visual imaging model; combining the estimated initial value of the illumination component and the distribution characteristics of a reflection component in actual imaging under the new visual imaging model to propose an optimization objective function under a variational framework; solving the proposed optimization objective function by using a variational optimization method to obtain an optimized illumination component and reflection component through decomposition; performing dynamic range adjustment on the illumination component obtained through decomposition of the optimization objective function; and mapping the optimized reflection component back to the illumination component with adjusted dynamic range according to the new visual imaging model, so that the low-light image achieves the enhancement effect of high visibility and high signal-to-noise ratio. The application can generate a low-light enhanced image with high visibility and high signal-to-noise ratio.

Description

TECHNICAL FIELD

[0001] The present application relates to a low-light image enhancement method, system, terminal and storage medium. BACKGROUND

[0002] Enhancing indoor and outdoor low-light scene images can improve the visual effect of the images, for example, many computer vision scenes, such as city intelligent management, intelligent video monitoring, face recognition systems, etc., have a high requirement for high-precision detection of effective features in images. However, in a low-light scene, due to the low light intensity in the natural environment and the widespread existence of scattering phenomena of small-form sol particles, dust and other particles in the atmosphere, the captured images have serious degradation problems, such as high noise, low visibility, color cast, etc. These problems greatly limit the application and landing of computer vision tasks.

[0003] Existing image enhancement techniques mainly include three categories: traditional enhancement methods, deep learning methods and Retinex model-based theoretical methods.

[0004] Traditional enhancement methods aim to adjust the dynamic range (contrast) by mapping the histogram to achieve visual enhancement of the target image. Although this method is simple to operate and has low computational cost, it emphasizes the enhancement of contrast rather than the adjustment of real light. Therefore, this method cannot take care of the relatively bright area pixels close to saturation or the relatively dark area pixels with very low intensity, and thus cannot better solve the ill-posed problem in the image enhancement process.

[0005] Compared with the former, the deep learning method has made great progress in enhancing images by using a large number of training samples, which can use big data samples and mapping constraints to solve the ill-posed problem in the enhancement process to a certain extent. However, this method relies on complex network structure and a large amount of training data, resulting in the problem of time-consuming training process and dependence on hardware resources. In addition, due to the difficulty of sample acquisition, when the test image is significantly different from the training image, the effect of solving the ill-posed problem by the deep learning method will decrease due to the weakening of the generalization ability, resulting in a decrease in precision.

[0006] Different from the first two methods, the basis of the Retinex model theory is that the color of an object is determined by the reflection ability of the object to long-wave (red), medium-wave (green) and short-wave (blue) light, rather than the absolute value of the reflected light intensity, and the color of the object is not affected by the non-uniformity of light, and has consistency, that is, Retinex is based on color consistency (color constancy). Different from the traditional linear, nonlinear method which can only enhance certain features of the image, Retinex can achieve a balance in dynamic range compression, edge enhancement and color constancy, so it can adaptively enhance various types of images. The estimated illumination component of this method is based on the assumption that the distribution of light in the actual scene is slowly smooth, and most of it is obtained by Gaussian filtering on the original image. Although this method can improve the visibility of brightness to some extent, the inaccurate illumination component estimated by the model due to unreasonable assumptions, and the assumption distribution of the reflection component and the noise component, result in visual artifacts or amplified noise in the output results, and finally the enhanced results have poor visibility and low signal-to-noise ratio.

[0007] Therefore, how to use a more scientific and reasonable mathematical model to construct the human eye visual imaging process and apply it to the low-light image enhancement task, so that the fine-processed image can obtain a clear view while meeting the demand of high-precision visual algorithm for high signal-to-noise ratio input, has become a research hotspot in the academic field and a rigid demand in the industrial field. SUMMARY

[0008] Therefore, it is necessary to provide a low-light image enhancement method to realize high-visibility and high-signal-to-noise ratio low-light image enhancement through iteration of the variational method.

[0009] The present application provides a low-light image enhancement method, which comprises the following steps: a. establishing a new visual imaging model according to the traditional Retinex visual imaging model; b. estimating the initial value of the illumination component according to the established new visual imaging model; c. under the new visual imaging model, combining the estimated initial value of the illumination component and the distribution characteristics of the reflection component in the actual imaging, and proposing an optimization objective function under the variational framework; d. solving the proposed optimization objective function by using the variational optimization method, and decomposing to obtain the optimized illumination component and reflection component; e. adjusting the dynamic range of the illumination component obtained by decomposing the optimization objective function by using a nonlinear gamma function; and mapping the optimized reflection component back to the illumination component with adjusted dynamic range according to the new visual imaging model, so that the low-light image achieves the enhancement effect of high visibility and high signal-to-noise ratio.

[0010] Preferably, in step a, the new visual imaging model is:

[0011] The new visual imaging model is:

[0012]

[0013] where P is the camera captured image, R is the reflectance component, L is the illumination component, is a pixel-wise pointwise operator, N is the noise component.

[0014] Preferably, the step b comprises:

[0015] In the three channels of the camera captured image P, the maximum value of the pixels between the channels is selected as the initial value of the illumination component, i.e. the initial value of the illumination component at each pixel x is: ini (x) = maxc∈{R,G,B}P(x).

[0016] Preferably, the step c comprises:

[0017] S3.1, the illumination component L satisfies the spatial piecewise smoothness in the whole image;

[0018] S3.2, the intensity value of the illumination component L in the whole image is greater than or equal to the intensity value of the captured image, i.e. L≥P;

[0019] S3.3, the reflectance component R satisfies the spatial piecewise constant in the whole image, and contains noise at low brightness;

[0020] S3.4, the details of the reflectance component R are consistent with the details in the captured image;

[0021] S3.5, it can be observed by experiments that: in a low light environment, the product of the reflectance component and the illumination component is close to the captured image, and the numerical proximity of the two is used to construct a corresponding data fidelity term;

[0022] By the above prior assumptions, an optimization objective function is proposed:

[0023]

[0024] where λ P , λ R , λ L are positive weight factors for balancing different regularization terms; is a data fidelity term, which is used to ensure that the recombined result is close to the captured image P; is a gradient fidelity term, which is used to keep the separated reflectance component R consistent with the captured image P in details; is used to force the reflectance component to keep spatial anisotropic smoothness, where acting on R to preserve important structures in non-homogeneous regions while compressing noise in homogeneous regions to improve signal-to-noise ratio; for enforcing sparsity and piecewise smoothness of the illumination component L.

[0025] Preferably, the step d comprises:

[0026] First, introduce two auxiliary variables and Then, the proposed optimization objective function is converted to the following conditional constrained optimization form:

[0027]

[0028]

[0029] The augmented Lagrangian function of the above equation is naturally written as:

[0030]

[0031] where ξ and η are the Lagrange multipliers, β L and β R are predefined positive parameters, and L A is separated into sub-problems corresponding to X and Y, L and R, and is solved alternately, converging to a global optimal solution for the next sub-problems;

[0032] (X, Y) sub-problem: Collect X and Y related terms from L A respectively, and obtain the minimization problem about X and Y in the following form:

[0033]

[0034]

[0035] The corresponding solution is:

[0036]

[0037]

[0038] where sgn represents the sign function;

[0039] (L, R) sub-problem: Fix the above X and Y results, collect L and R related terms from L A respectively, and obtain the minimization problem about L and R in the following form:

[0040]

[0041]

[0042] Its corresponding solution is:

[0043]

[0044]

[0045] where I denotes the identity matrix, F(·), F -1 (·) and denote the FFT operator, the inverse FFT operator and the complex conjugate operator, respectively;

[0046] (ξ,η) update: in each iteration, the Lagrange multipliers ξ and η are updated using step size and , respectively;

[0047] Based on the above operations, X and Y, L and R obtain the optimal solution in the iteration and obtain the optimized illumination component and the low-noise and sharpened reflection component.

[0048] Preferably, the step e comprises:

[0049] The illumination component and the reflection component decomposed in step d are denoted as L a and R a , respectively, the dynamic range of the illumination component is nonlinearly adjusted using gamma = 1 / 2.2 to obtain the enhancement of the illumination component in brightness, and the adjusted result is denoted as L' a ;

[0050] The low-noise and sharpened reflection component R a decomposed in step d is mapped back to the illumination component L' a adjusted in the dynamic range to obtain the enhanced output

[0051] ​The application further provides a low-light image enhancement system, comprising: a model establishing module, configured to establish a new visual imaging model according to a traditional Retinex visual imaging model; an initial value estimating module, configured to estimate an initial value of an illumination component according to the established new visual imaging model; a function proposing module, configured to propose an optimization objective function under a variational framework by combining the estimated initial value of the illumination component and distribution characteristics of a reflection component in actual imaging under the new visual imaging model; a function solving module, configured to solve the proposed optimization objective function by using a variational optimization method, and decompose to obtain an optimized illumination component and reflection component; and a dynamic adjustment module, configured to perform dynamic range adjustment on the illumination component obtained by decomposing the optimization objective function by using a nonlinear gamma function; and according to the new visual imaging model, map the optimized reflection component back to the illumination component with adjusted dynamic range, so that the low-light image achieves an enhancement effect of high visibility and high signal-to-noise ratio.

[0052] The application further provides a terminal, comprising a processor and a memory coupled to the processor, wherein: the memory stores program instructions for implementing the low-light image enhancement method; and the processor is configured to execute the program instructions stored in the memory to implement low-light image enhancement.

[0053] The application further provides a storage medium storing processor-executable program instructions for executing the low-light image enhancement method.

[0054] The application analyzes the reason for low signal-to-noise ratio enhancement by introducing a noise superposition model ignored by a traditional Retinex model, and reasonably assumes the illumination component and the reflection component to propose a new optimization objective function under a variational framework. Then, the optimization objective function is solved and accelerated by using an iterative optimization and fast Fourier method. Finally, a low-light enhanced image with high visibility and high signal-to-noise ratio is finally generated by combining the optimized illumination component and the low-noise reflection component. In addition, the application can be extended to handle a series of similar problems, such as underwater or remote sensing image enhancement, and image restoration under conditions such as haze or dust.

[0055] The beneficial effects of the application include:

[0056] Firstly, the low-light imaging process is modeled based on a new form of visual imaging model. An optimization objective function capable of decomposing and coupling the reflection component and the illumination component is constructed under a variational framework;

[0057] Secondly, the constructed objective function can not only separate the piecewise smooth illumination component, but also compress noise while ensuring the sharpness of the reflection component, thereby improving the signal-to-noise ratio;

[0058] Thirdly, the application can realize real-time processing effect by using a fast solving method and a high-performance platform, and is suitable for being embedded into low-light image enhancement processing of an embedded development platform, and has very wide application prospects. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 A flow chart of the low-light image enhancement method of the application is shown in the figure.

[0060] Figure 2 A low-light image enhancement process schematic diagram provided by the embodiment of the application is shown in the figure, wherein (a) is a low-light image, (b) is an initial illumination map, (c) is an optimized decomposed illumination map, (d) is an adjusted illumination map, (e) is an optimized decomposed reflection component, and (f) is an enhanced effect diagram.

[0061] Figure 3 A result comparison schematic diagram of different low-light image enhancement methods on an LIME data set provided by the embodiment of the application is shown in the figure.

[0062] Figure 4 A result comparison schematic diagram of different low-light image enhancement methods on an LOL data set provided by the embodiment of the application is shown in the figure.

[0063] Figure 5 A structure schematic diagram of the low-light image enhancement system of the embodiment of the application is shown in the figure.

[0064] Figure 6 A terminal structure schematic diagram of the embodiment of the application is shown in the figure.

[0065] Figure 7 A structure schematic diagram of the storage medium of the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0066] The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] Reference Figure 1 is a work flow chart of the preferred embodiment of the low-light image enhancement method of the application.

[0068] Step S1, according to a traditional Retinex visual imaging model, a new visual imaging model is established.

[0069] Specifically,

[0070] The traditional Retinex visual imaging model is:

[0071]

[0072] where P is the camera captured image, R is the reflectance component, L is the illumination component, and o is a pixel-wise pointwise operator. This model does not fully represent the imaging process in the actual noise scenario, and thus, the present application adjusts the traditional Retinex visual imaging model to the following new form:

[0073]

[0074] where N is the noise component. The visual imaging model in the new form (i.e., the new visual imaging model) can be intuitively explained in the physical sense as follows: in a low-light environment, the noise attached to the reflectance component is the reason for the low signal-to-noise ratio in the final enhanced result. Thus, how to construct a new optimization objective to make the separated reflectance component consistent with the input image in the details while compressing the noise becomes a key.

[0075] Step S2, estimating the initial value of the illumination component according to the established new visual imaging model. That is, taking the maximum value of the pixels among the three channels in the camera captured image as the initial value of the estimated illumination component. Specifically:

[0076] In the three channels of the above camera captured image P, the maximum value of the pixels among the channels is selected as the initial value of the illumination component, that is, the initial value of the illumination component at each pixel x is: ini (x)=maxc∈{R,G,B}P(x).

[0077] Step S3, prior assumption of the model component and design of the optimization objective function. That is, under the new visual imaging model, combining the estimated initial value of the illumination component and the distribution characteristics of the reflectance component in the actual imaging, an optimization objective function under the variational framework is proposed. The optimization objective function can separate the illumination component with piecewise smoothness and the reflectance component with piecewise constant and noise compression. Specifically:

[0078] The present embodiment makes the following prior assumptions and constraints on the illumination component and the reflectance component according to the new visual imaging model proposed by the present application:

[0079] S3.1, the illumination component L satisfies the spatial piecewise smoothness in the entire image;

[0080] S3.2, the intensity value of the illumination component L in the entire image is greater than or equal to the intensity value of the captured image, that is, L≥P;

[0081] S3.3, the reflectance component R satisfies the spatial piecewise constant in the entire image, and contains noise at low brightness;

[0082] S3.4, the details of the reflectance component R are consistent with the details in the captured image;

[0083] S3.5, it can be observed by experiments that: in low light environment, the product of the reflection component and the illumination component is close to the captured image, for the numerical similarity of the two (which can be approximately understood as: to construct the corresponding data fidelity term;

[0084] By the above prior assumptions, the following optimization objective function is proposed:

[0085]

[0086] wherein λ P , λ R , λ L are positive weight factors for balancing different regularization terms, which are empirically set to 0.001, 0.001, 0.01 respectively; is the data fidelity term, which is used to ensure that the recombined result is close to the captured image P; is the gradient fidelity term, which is used to keep the separated reflection component R consistent with the captured image P in details; is used to force the reflection component to keep spatial anisotropic smoothing, wherein acts on R to retain important structures in non-homogeneous regions while compressing noise in homogeneous regions to improve the signal-to-noise ratio, where τ = 10 -5 to prevent the denominator from being 0; is used to force the illumination component L to be sparse and piecewise smooth, wherein the weight with control parameter γ can distinguish between keeping the depth discontinuity edge features and smoothing the texture of homogeneous regions, where B is the bright channel of the captured image, i.e. B = L ini ; is the first-order difference, which contains the vertical direction and the horizontal direction

[0087] By a series of prior constraint conditions, the human eye vision model in low light environment is processed, and the optimization objective function for the illumination component and the low-noise and sharpened reflection component is constructed in the variational framework. The significance of this process in the embodiment is to effectively convert the original ill-posed problem into a solvable well-posed problem.

[0088] Step S4, solving the optimization objective function: the variational optimization method is used to solve the proposed optimization objective function, and the optimized illumination component and reflection component are obtained by decomposition. Specifically:

[0089] After qualitative analysis, under the trade-off between the data fidelity term and the constraint term, the following conclusions are drawn, namely:

[0090] The estimate of the illumination component L should exhibit the following properties: a) piecewise smooth behavior, and preserving the depth discontinuity at abrupt edges in the spatial scene; b) it should have higher intensity values than the captured image.

[0091] While the estimate of the reflectance component R should exhibit the following properties: a) piecewise constant behavior, and preserving the depth continuity at abrupt edges in the spatial scene; b) the amount of noise contained should be compressed, thus improving the signal-to-noise ratio; c) it should be consistent with the captured image in the details representation, thus enhancing the sharpening effect of the enhanced result.

[0092] Based on this conclusion, a variety of fast solving algorithms are used to solve the optimization objective function proposed in the present application. The present embodiment adopts an estimation method based on fast Fourier transform to solve the illumination component and the reflectance component, and other solving schemes such as QR fast decomposition, low-rank space conversion, etc. can also be used to achieve the same. Thus, as long as the solving method of the optimization objective function constructed by the variational optimization method is within the protection scope of the present application.

[0093] Considering the non-smoothness penalty of the L1 norm in the optimization objective function proposed, it causes difficulty in calculation. Therefore, the present embodiment proposes to use the alternating direction minimization technique to solve the optimization objective function proposed. Here, two auxiliary variables and are first introduced, and then the optimization objective function proposed is converted into the following optimization form with conditional restrictions:

[0094]

[0095]

[0096] The augmented Lagrangian function of the above formula can be naturally written as:

[0097]

[0098] wherein, ξ and η are Lagrange multipliers, and β L and β R are predefined positive parameters (both are empirically set to 1). Next, L A is separated into sub-problems corresponding to X and Y, L and R, and solved alternately for the convergence of the following sub-problems to a global optimal solution.

[0099] (X, Y) sub-problem: X and Y related terms are collected from L A respectively, and the minimization problem about X and Y is obtained in the following form:

[0100]

[0101]

[0102] The corresponding solution is:

[0103]

[0104]

[0105] where sgn denotes the sign function.

[0106] (L, R) sub-problem: Fix the results of X and Y above, collect L and R related terms from L A and R respectively, and obtain the minimization problem about L and R in the following form:

[0107]

[0108]

[0109] The corresponding solution is:

[0110]

[0111]

[0112] where I denotes the identity matrix, F(·), F -1 (·) and denote the FFT operator, the inverse FFT operator and the complex conjugate operator respectively.

[0113] (ξ, η) update: In each iteration process, the Lagrange multipliers ξ and η can be updated by and respectively.

[0114] Based on the above operations, X and Y, L and R obtain the optimal solution in the iteration and obtain the optimized illumination component and the low-noise and sharpened reflection component.

[0115] Step S5, output of high visibility and high signal-to-noise ratio image, that is, dynamic range adjustment of the illumination component obtained by decomposing the optimization objective function by using a nonlinear gamma function; according to the new visual imaging model, the optimized reflection component is mapped back to the illumination component with adjusted dynamic range, so that the low-light image achieves the enhancement effect of high visibility and high signal-to-noise ratio. Specifically:

[0116] S5.1, according to the illumination component and the reflection component decomposed in step S4, they are denoted as L a and R a respectively, as shown in Figure 2 (c) and Figure 2 ​(e) as shown. Then, the dynamic range of the illumination component is nonlinearly adjusted by gamma = 1 / 2.2 to obtain the enhancement of the illumination component in luminance, and the adjusted result is recorded as L' a , as shown in (d); Figure 2 (d) as shown;

[0117] S5.2, the low-noise and sharpened reflection component R a mapped back to the illumination component L' of the adjusted dynamic range to obtain the enhanced output a as shown in (f). From the enhancement effect, the noise is greatly compressed, and a pleasing visual performance is obtained. as shown in (f). From the enhancement effect, the noise is greatly compressed, and a pleasing visual performance is obtained. Figure 2 (f) as shown. From the enhancement effect, the noise is greatly compressed, and a pleasing visual performance is obtained.

[0118] Referring to Figure 5 , which is a hardware architecture diagram of the low-light image enhancement system 10 of the present application. The system includes a model establishing module 101, an initial value estimation module 102, a function proposing module 103, a function solving module 104, and a dynamic adjustment module 105. Among them:

[0119] The model establishing module 101 is used to establish a new visual imaging model according to the traditional Retinex visual imaging model;

[0120] The initial value estimation module 102 is used to estimate the initial value of the illumination component according to the established new visual imaging model;

[0121] The function proposing module 103 is used to propose an optimization objective function under the variational framework, in combination with the estimated initial value of the illumination component and the distribution characteristics of the reflection component in actual imaging, under the new visual imaging model.

[0122] Among them, the optimization objective function can separate out the illumination component with piecewise smoothness and the reflection component with piecewise constant and noise compression.

[0123] The function solving module 104 is used to solve the proposed optimization objective function by using the variational optimization method, and to obtain the optimized illumination component and reflection component by decomposition;

[0124] The dynamic adjustment module 105 is used to adjust the dynamic range of the illumination component obtained by the optimization objective function by using a nonlinear gamma function; according to the new visual imaging model, the optimized reflection component is mapped back to the illumination component of the adjusted dynamic range, so that the low-light image achieves the enhancement effect of high visuality and high signal-to-noise ratio.

[0125] Please refer to Figure 6 , which is a terminal structure diagram of the embodiment of the present application. The terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51.

[0126] The memory 52 stores program instructions for implementing the low-light image enhancement method described above.

[0127] The processor 51 is configured to execute the program instructions stored in the memory 52 to implement the low-light image enhancement.

[0128] The processor 51 can also be referred to as a CPU (Central Processing Unit). The processor 51 can be an integrated circuit chip having signal processing capability. The processor 51 can also be a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application-Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0129] Referring to Figure 7 , it is a structural schematic diagram of the storage medium of the embodiment of the present application. The storage medium of the embodiment of the present application stores a program file 61 capable of implementing all the methods described above, wherein the program file 61 can be stored in the storage medium in the form of a software product, including a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk or an optical disk, and various media capable of storing program codes, or a terminal device such as a computer, a server, a mobile phone, a tablet, etc.

[0130] The present application proposes an optimization objective function under a variational framework by analyzing and assuming the distribution characteristics of actual illumination components and reflection components. Finally, the optimization objective function is solved by an alternating direction minimization method and a fast Fourier transform method, and a low-light image enhancement result with high visibility and high signal-to-noise ratio is generated (as shown in Table I). Since the SRIE algorithm and the LIME algorithm have unreasonable assumptions about the components in the assumptions of the model, artifacts and overexposure problems will occur in local regions (as shown in Figure 3 (c) and Figure 3 (d), which leads to the problems of undesirable visual effects and low signal-to-noise ratio in the enhancement results of the two algorithms, such as Figure 4 (d) and Figure 4(e) are shown. In comparison, the present application makes reasonable assumptions on the actual distribution of illumination model and reflection model based on a reasonable visual imaging model, and respectively enhances and suppresses the details and noise in the components, so as to achieve a pleasing visual performance in brightness, contrast, and detail sharpness, while achieving a high signal-to-noise ratio on the basis of compressed noise.

[0131] Table I. Average SSIM and PSNR results of different enhancement methods on 500 image pairs (low-light image and reference image) from the LOL dataset

[0132]

[0133] Although the present application is described with reference to the current preferred embodiments, those skilled in the art will understand that the above-described embodiments are intended to be illustrative only and not limiting of the scope of the application, which is set forth in the claims. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the scope of protection of the present application.

Claims

1. A low-light image enhancement method, characterized by, The method comprises the following steps: a. establishing a new visual imaging model according to a traditional Retinex visual imaging model; b. estimating an initial value of the illumination component according to the established new visual imaging model; c. under the new visual imaging model, combining the estimated initial value of the illumination component and the distribution characteristics of the reflection component in actual imaging, and proposing an optimization objective function under a variational framework; d. solving the proposed optimization objective function by using a variational optimization method to obtain the optimized illumination component and reflection component through decomposition; e. adjusting the dynamic range of the illumination component obtained through decomposition of the optimization objective function by using a nonlinear gamma function; and mapping the optimized reflection component back to the illumination component with adjusted dynamic range according to the new visual imaging model, so that the low-light image achieves the enhancement effect of high visibility and high signal-to-noise ratio; wherein: in the step a: the new visual imaging model is: where P is the camera captured image, R is the reflectance component, L is the illumination component, is a pixel-based pointwise multiplication operator, and N is a noise component. the step c comprises: S3.1, the illumination component L satisfies the spatial piecewise smoothness in the whole image; S3.2, the intensity value of the illumination component L within the entire image is greater than or equal to the intensity value of the captured image, i.e. L > 0 P ; S3.3, the reflection component R satisfies the spatial piecewise constant and contains noise at low brightness in the whole image; S3.4, the details of the reflection component R are consistent with the details in the captured image; S3.5, it can be observed through experiments that: under a low-light environment, the product of the reflection component and the illumination component is close to the captured image, and a corresponding data fidelity term is constructed for the numerical proximity of the two; through the above prior assumptions, an optimization objective function is proposed: where λ P , λ R , λ L are positive weight factors for balancing different regularizers; is a data fidelity term for ensuring that the recombined result is close to the captured image P; is a gradient fidelity term for keeping the separated reflection component R consistent with the captured image P in details; is for enforcing the reflection component to keep spatially anisotropic smoothing, where acts on R to preserve important structures in non-homogeneous regions while compressing noise in homogeneous regions to improve signal-to-noise ratio; is for enforcing the illumination component L to be sparse and piecewise smooth.

2. The method of claim 1, wherein, the step b comprises: In the three channels of the camera capturing the image P, the maximum value of the pixels between the channels is selected as the initial value of the illumination component, that is, the initial value of the illumination component at each pixel x is: ini (x) = max c∈{R,G,B} P(x).

3. The method of claim 2, wherein, the step d comprises: First, two auxiliary variables are introduced and The proposed optimization objective function is then converted to the following form of optimization with conditional constraints: The augmented Lagrange function of the above formula is naturally written as: where ξ and η are Lagrange multipliers, β L and β R are predefined positive parameters, and L A is split into subproblems corresponding to X and Y, L and R and solved alternately for convergence to a global optimal solution for the next subproblem; (X, Y) subproblem: Collect X and Y related terms from L A respectively, and obtain the minimization problem about X and Y in the following form: the corresponding solution is: wherein, sgn represents a sign function; ( L Sub-problem: Fix X and Y from above, collect L and R terms from L A and R respectively, and obtain minimization problems for L and R in the form: the corresponding solution is: where I denotes an identity matrix, F(·), F -1 (·) and denote an FFT operation symbol, an inverse FFT operator and a complex conjugate operator, respectively; (ξ, η) update: At each iteration, the Lagrange multipliers ξ and η are updated using step sizes of respectively by and . Based on the above operation, X and Y, L and R obtain the optimal solution in iteration and obtain the optimized illumination component and the low-noise and sharpened reflection component.

4. The method of claim 3, wherein, the step e comprises: The illumination component and the reflection component decomposed from step d are denoted as L a and R a respectively. The dynamic range of the illumination component is adjusted non-linearly using gamma = 1 / 2.2 to obtain the enhancement of the illumination component in luminance, and the adjusted result is denoted as L′ a ; the low-noise and sharpened reflection component R broken out of step d a mapping the light component L' back to an adjusted dynamic range a the above to obtain an enhanced output 5. A low-light image enhancement system using the low-light image enhancement method of claim 1, characterized by, including: a model establishing module for establishing a new visual imaging model according to a traditional Retinex visual imaging model; an initial value estimating module for estimating an initial value of the illumination component according to the established new visual imaging model; a function proposing module for proposing an optimization objective function under a variational framework by combining the estimated initial value of the illumination component and the distribution characteristics of the reflection component in actual imaging under the new visual imaging model; a function solving module for solving the proposed optimization objective function by using a variational optimization method to obtain the optimized illumination component and reflection component through decomposition; a dynamic adjusting module for adjusting the dynamic range of the illumination component obtained through decomposition of the optimization objective function by using a nonlinear gamma function; and mapping the optimized reflection component back to the illumination component with adjusted dynamic range according to the new visual imaging model, so that the low-light image achieves the enhancement effect of high visibility and high signal-to-noise ratio.

6. A terminal, characterized by comprising: The terminal comprises a processor and a memory coupled with the processor, wherein: the memory stores program instructions for implementing the low-light image enhancement method of any one of claims 1-4; the processor is configured to execute the program instructions stored in the memory to implement the low-light image enhancement.

7. A storage medium, characterized by A computer program product comprising a computer readable medium having stored thereon processor-executable program instructions for causing a processor to perform the low-light image enhancement method according to any one of claims 1 to 4. A computer program product comprising a computer readable medium having stored thereon processor-executable program instructions for causing a processor to perform the low-light image enhancement method according to any one of claims

Images