A low-illumination image enhancement method, system, device and storage medium
By using an integrated three-channel image enhancement network model, preprocessing, image regression, and local adjustment are performed. Inverse illumination map data is calculated and Hadamard product is executed, which solves the problem of uneven exposure in low-light images and achieves efficient and uniform image enhancement.
Patent Information
- Application Number
- CN202211702476.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-28
- Publication Date
- 2026-03-03
- Estimated Expiration
- 2042-12-28
AI Technical Summary
Existing technologies struggle to effectively address low-light images that are both overexposed and underexposed, failing to maintain image contrast, naturalness, and object detail.
An integrated three-channel image enhancement network model is adopted, including an image preprocessing network, a first processing channel, a second processing channel, and a third processing channel. Through preprocessing, image regression, global enhancement, and local adjustment, inverse illumination map data is calculated and Hadamard product is performed to achieve image enhancement.
It achieves efficient and uniform low-light image enhancement, solves the problem of uneven image illumination distribution, saves computational costs, and has good robustness.
Smart Images
Figure CN115861125B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of image processing technology, and specifically relates to a low-light image enhancement method, system, device, and storage medium. Background Technology
[0002] Low-light image enhancement methods can be divided into traditional methods and deep learning-based methods. Among traditional methods, histogram equalization (HEM) and Retinex model-based methods are representative. Histogram equalization methods approximate a uniform pixel distribution histogram of the image to be enhanced, thereby improving image contrast. Retinex model-based methods assume that the image seen by the human eye consists of illumination and reflection components; the illumination component changes with lighting conditions, while the reflection component remains constant. These algorithms primarily enhance the image by estimating the reflection component of the image to be enhanced. However, in practice, these traditional methods only achieve certain results in simple lighting environments with low noise interference. Real-world enhancement scenarios are often complex and noisy, and the limitations of traditional methods make them unsuitable for these more complex scenarios.
[0003] Low-light image enhancement methods based on deep learning can be divided into two categories: CNN (Convolutional Neural Networks) and GAN (Generative Adversarial Networks). While existing deep learning-based methods have to some extent compensated for the shortcomings of traditional methods, the main challenge in low-light image enhancement is how to suppress overexposed areas and enhance underexposed areas in images that are both overexposed and underexposed, while simultaneously maintaining image contrast, naturalness, and object details. Current deep learning-based methods still cannot effectively solve this challenge. Summary of the Invention
[0004] The purpose of this invention is to provide a low-light image enhancement method, system, device, and storage medium to solve the aforementioned problems existing in the prior art.
[0005] To achieve the above objectives, the present invention adopts the following technical solution:
[0006] Firstly, a low-light image enhancement method is provided, comprising:
[0007] Acquire low-light image data of the target;
[0008] The target low-light image data is imported into a pre-set image enhancement network model for image enhancement processing to obtain weight parameters, global enhancement map data, and local adjustment map data. The image enhancement network model includes an image preprocessing network, a first processing channel, a second processing channel, and a third processing channel. The image preprocessing network is used to preprocess the target low-light image data to extract several feature map data. The first processing channel is used to perform image regression processing on the several feature map data to obtain weight parameters. The second processing channel is used to perform global enhancement processing on the several feature map data to obtain global enhancement map data. The third processing channel is used to perform local adjustment processing on the several feature map data to obtain local adjustment map data.
[0009] Inverse illumination map data is calculated based on weight parameters, global enhancement map data, and local adjustment map data;
[0010] The inverse illumination map data and the target low-light image data are subjected to a Hadamard product operation to obtain the target low-light image enhanced image data.
[0011] In one possible design, the preprocessing network uses the ResNet50 model to preprocess the target low-light image data to extract several feature map data.
[0012] In one possible design, the first processing channel includes a regression module comprising a convolutional kernel, an average pooling kernel, a multilayer perceptron, and a sigmoid layer connected in sequence, with the sigmoid layer outputting weight parameters in the range (0,1).
[0013] In one possible design, the second processing channel includes a straightening module, a position encoding module, a Swing Transformer module, and a first Task Tail Block module connected in sequence. The straightening module is used to straighten the feature map data. The position encoding module is used to perform position encoding on the output of the straightening module to obtain the first tokens parameter. The Swing Transformer module is used to input the first tokens parameter to perform global modeling and multi-scale feature extraction, and output the second tokens parameter. The first Task Tail Block module is used to input the second tokens parameter to perform linear mapping and rearrangement processing, and output global enhanced map data.
[0014] In one possible design, the third processing channel includes a second Task Tail Block module, which is used to input several feature map data for linear mapping and rearrangement processing, and output local adjustment map data.
[0015] In one possible design, the loss function used by the image enhancement network model is:
[0016] L total =L spa +λ1L exp +λ2L col +λ3L smo
[0017] Among them, L total Let L be the overall loss function of the image enhancement network model. spa Characteristic space consistency loss function, L exp Characterizing the exposure control loss function, L col The color consistency loss function, L smo The loss function characterizing illuminance smoothness, λ1, λ2, and λ3 are respectively L exp L col and L smo The weight.
[0018] In one possible design, the formula for calculating the inverse illumination map data based on the weight parameters, global enhancement map data, and local adjustment map data is as follows:
[0019]
[0020] in, For inverse illumination map data, For globally enhanced graph data, This is data for a locally adjusted plot, where α is the weighting parameter.
[0021] The formula for calculating the Hadamard product of the inverse illumination map data and the target low-light image data is as follows:
[0022]
[0023] Where R represents the enhanced low-light image data of the target, and L represents the low-light image data of the target. Characterizes the Hadamard product operator.
[0024] Secondly, a low-light image enhancement system is provided, comprising an acquisition unit, a processing unit, a first calculation unit, and a second calculation unit, wherein:
[0025] The acquisition unit is used to acquire low-light image data of the target.
[0026] The processing unit is used to import the target low-light image data into a preset image enhancement network model for image enhancement processing to obtain weight parameters, global enhancement map data, and local adjustment map data. The image enhancement network model includes an image preprocessing network, a first processing channel, a second processing channel, and a third processing channel. The image preprocessing network is used to preprocess the target low-light image data to extract several feature map data. The first processing channel is used to perform image regression processing on the several feature map data to obtain weight parameters. The second processing channel is used to perform global enhancement processing on the several feature map data to obtain global enhancement map data. The third processing channel is used to perform local adjustment processing on the several feature map data to obtain local adjustment map data.
[0027] The first calculation unit is used to calculate the inverse illumination map data based on the weight parameters, global enhancement map data, and local adjustment map data;
[0028] The second calculation unit is used to perform a Hadamard product operation on the inverse illumination map data and the target low-light image data to obtain the enhanced image data of the target low-light image.
[0029] Thirdly, a low-light image enhancement device is provided, comprising:
[0030] Memory, used to store instructions;
[0031] A processor is configured to read instructions stored in the memory and execute the method described in any one of the first aspects above, according to the instructions.
[0032] Fourthly, a computer-readable storage medium is provided, on which instructions are stored, which, when executed on a computer, cause the computer to perform any of the methods described in the first aspect. A computer program product containing instructions is also provided, which, when executed on a computer, cause the computer to perform any of the methods described in the first aspect.
[0033] Beneficial Effects: This invention enhances low-light image data by importing it into a "one-in-one three-channel" image enhancement network model, obtaining weight parameters, global enhancement map data, and local adjustment map data. Then, inverse illumination map data is calculated based on these data. Finally, the inverse illumination map data is combined with the target low-light image data using a Hadamard product to obtain the enhanced image data. This achieves efficient and uniform low-light image enhancement. This invention employs a global enhancement and local adjustment approach, effectively solving the problem of difficult low-light image enhancement caused by uneven image illumination distribution. It creatively utilizes a "one-in-one three-channel" image enhancement network structure to extract global and local features, saving computational costs. Furthermore, by uniformly handling both uniform and non-uniform image illumination through weight parameters, it exhibits excellent robustness. Attached Figure Description
[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0035] Figure 1 This is a schematic diagram of the steps of the method in an embodiment of the present invention;
[0036] Figure 2 This is a schematic diagram of the processing flow of the image enhancement network model in an embodiment of the present invention;
[0037] Figure 3 This is a schematic diagram of the Regression module in an embodiment of the present invention;
[0038] Figure 4 This is a schematic diagram of the Swin Transformer module in an embodiment of the present invention;
[0039] Figure 5 This is a schematic diagram of the structure of two Task Tail Block modules in an embodiment of the present invention;
[0040] Figure 6 This is a schematic diagram of the system configuration in an embodiment of the present invention;
[0041] Figure 7 This is a schematic diagram of the device configuration in an embodiment of the present invention. Detailed Implementation
[0042] It should be noted that the descriptions of these embodiments are intended to aid in understanding the invention and do not constitute a limitation thereof. The specific structural and functional details disclosed herein are merely for describing exemplary embodiments of the invention. However, the invention may be embodied in many alternative forms and should not be construed as being limited to the embodiments described herein.
[0043] It should be understood that, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments according to the specific circumstances.
[0044] Specific details are provided in the following description to provide a complete understanding of the exemplary embodiments. However, those skilled in the art will understand that the exemplary embodiments can be implemented without these specific details. For example, the system may be shown in block diagrams to avoid obscuring the example with unnecessary details. In other embodiments, well-known processes, structures, and techniques may be shown without non-essential details to avoid obscuring the embodiments.
[0045] Example 1:
[0046] This embodiment provides a low-light image enhancement method, which can be applied to corresponding image processing servers, such as... Figure 1 and Figure 2 As shown, the method includes the following steps:
[0047] S1. Acquire low-light image data of the target.
[0048] In practice, the image processing server first obtains the target low-light image data that needs to be enhanced. For example, the size of the target low-light image can be set to 224×224×3.
[0049] S2. The target low-light image data is imported into a preset image enhancement network model for image enhancement processing to obtain weight parameters, global enhancement map data, and local adjustment map data. The image enhancement network model includes an image preprocessing network, a first processing channel, a second processing channel, and a third processing channel. The image preprocessing network is used to preprocess the target low-light image data to extract several feature map data. The first processing channel is used to perform image regression processing on the several feature map data to obtain weight parameters. The second processing channel is used to perform global enhancement processing on the several feature map data to obtain global enhancement map data. The third processing channel is used to perform local adjustment processing on the several feature map data to obtain local adjustment map data.
[0050] In practice, after acquiring the target low-light image data, the image processing server imports the target low-light image data into a pre-set image enhancement network model for image enhancement processing, obtaining weight parameters, global enhancement map data, and local adjustment map data. The image enhancement network model includes an image preprocessing network, a first processing channel, a second processing channel, and a third processing channel. The image preprocessing network is used to preprocess the target low-light image data to extract several feature map data. The first processing channel is used to perform image regression processing on the several feature map data to obtain weight parameters. The second processing channel is used to perform global enhancement processing on the several feature map data to obtain global enhancement map data. The third processing channel is used to perform local adjustment processing on the several feature map data to obtain local adjustment map data.
[0051] The preprocessing network uses the ResNet50 model to preprocess the target low-light image data to extract several feature map data. For example, 2048 7×7 feature map data are obtained through the ResNet50 model.
[0052] The first processing channel includes a regression module, the structure of which is as follows: Figure 3 As shown, it includes a convolutional kernel, an average pooling kernel, a multilayer perceptron, and a sigmoid layer connected in sequence. The size of the convolutional kernel is 3×3, the stride is 2, the number of input channels is 2048, and the number of output channels is 256. The size of the average pooling kernel is 3×3, which is flattened to 1×256, and then passed through two multilayer perceptrons to become 1×1. The sigmoid activation function of the sigmoid layer outputs a weight parameter α in the range of (0,1).
[0053] The second processing channel includes a flattening module, a position encoding module, a Swin Transformer module, and a first Task Tail Block module connected in sequence. The flattening module is used to flatten the feature map data. The position encoding module is used to encode the output of the flattening module to obtain the first tokens parameter (i.e., 49×2048 tokens). The structure of the Swin Transformer module is as follows: Figure 4 As shown, the Swin Transformer module is used to input the first tokens parameter for global modeling and multi-scale feature extraction, and output the second tokens parameter (i.e., 196×128 tokens). The structure of the first Task Tail Block module is as follows. Figure 5 The left-hand side, Task Tail Block-1, includes a linear mapping layer and a rearrangement layer. The linear mapping layer has an input of 196×128 and an output of 196×768. The rearrangement layer outputs 224×224×3. The first Task Tail Block takes the second tokens parameter as input for linear mapping and rearrangement processing, and outputs 224×224×3 global augmentation graph data.
[0054] The third processing channel includes a second Task Tail Block module, the structure of which is the same as that of the first Task Tail Block module, such as... Figure 5 The right-hand side of Task Tail Block-2 is shown. It is used to input several feature map data, perform linear mapping and rearrangement processing, and output 224×224×3 local adjustment map data, i.e.
[0055] The loss function used in the image enhancement network model is:
[0056] L total =L spa +λ1L exp +λ2L col +λ3L smo
[0057] Among them, L total L represents the overall loss function of the image enhancement network model. spa The spatial consistency loss function primarily controls the correlation between adjacent pixels; L exp The exposure control loss function characterizes the image exposure level; L colThe color consistency loss function primarily adjusts the color correlation of the RGB three channels; L smo The illumination smoothness loss function is used to maintain the monotonicity of adjacent pixels; λ1, λ2, and λ3 are respectively L exp L col and L smo The weight.
[0058] Spatial consistency loss function L spa The coherence of the output image is enhanced by preserving the differences between neighboring regions of the input image and the enhanced image. The expression is:
[0059]
[0060] Where K is the number of local regions, Ω(i) is the set of the four adjacent regions (top, bottom, left, and right) of the selected region, and Y and I are the average illumination intensities of the corresponding regions in the output and input images, respectively. Experiments have verified that the size of the local region is set to 4×4.
[0061] To adjust underexposed or overexposed areas, an exposure control loss function L can be designed. exp Control the exposure level, the expression is:
[0062]
[0063] Exposure control loss function L exp The distance between the average intensity of a local area and the ideal exposure intensity E was measured. Based on this, experiments showed that setting E=0.62 usually yielded the best results. Here, M represents the number of non-overlapping local areas, and Y is the average intensity value of a local area in the enhanced image. Experiments verified that a local area size of 16×16 was optimal.
[0064] Color consistency loss function L col By establishing the relationship between the RGB three channels, potential color deviations in the enhanced image can be corrected. The expression is as follows:
[0065]
[0066] In the formula, J p q represents the average intensity of the p channel in the enhanced image, and (p,q) represents the pairwise combination of different channels.
[0067] Illuminance smoothness loss function L smo This is to control the monotonicity of adjacent pixels; the expression is:
[0068]
[0069] Where, M G The global augmentation graph, ML Refers to the local fine-tuning diagram. and These represent the gradients in the horizontal and vertical directions, respectively.
[0070] Experiments have shown that values of 2, 0.5, and 10 for λ1, λ2, and λ3, respectively, can achieve relatively ideal results.
[0071] S3. Calculate the inverse illumination map data based on the weight parameters, global enhancement map data, and local adjustment map data.
[0072] In practice, after processing by the image enhancement network model to obtain the weight parameter α and the global enhancement map data... and local adjustment map data Then, based on the weight parameter α and the global augmentation graph data, and local adjustment map data Calculate the inverse illumination map data The calculation formula is For a specific image, the degree of dependence between global and local adjustments is controlled by the weight parameter α.
[0073] S4. Perform a Hadamard product operation on the inverse illumination map data and the target low-light image data to obtain the enhanced target low-light image data.
[0074] In practice, after obtaining the inverse illumination map data, the Hadamard product operation can be performed between the inverse illumination map data and the target low-light image data based on Retinex theory to obtain the enhanced target low-light image data. The calculation formula is as follows:
[0075]
[0076] Where R represents the enhanced low-light image data of the target, and L represents the low-light image data of the target. Characterizes the Hadamard product operator, used for the target low-light image data L and inverse illumination map data in the Hadamard product operation. Given two matrices of the same size, the product of corresponding elements forms a new matrix, which is the enhanced image data.
[0077] Based on Retinex theory, low-light image modeling:
[0078]
[0079] L is the target low-light image, R is the ideal image to be recovered, and T represents the illumination map; let ε be a very small constant (approaching zero), and denote the matrix formed by the reciprocals of corresponding elements in T+ε as... but:
[0080]
[0081] In the formula, w is the width of the image, h is the height of the image, c is the number of channels of the image, and 1 w×h×c It is a matrix consisting entirely of 1s. Therefore:
[0082]
[0083] When the left side equals R, the right-hand term εR→0, therefore:
[0084]
[0085] This formula shows that the process of recovering an ideal image can be viewed as a process of inverse illumination of the original image, provided that the original illumination map T or the inverse illumination map is known. Therefore, this embodiment obtains the weight parameter α and global augmentation map data through image augmentation network model processing. and local adjustment map data Then, the inverse illumination map data was calculated. We can then utilize the inverse illumination map data. The ideal image is obtained by combining the original image data with the image data of the target low-light image.
[0086] Example 2:
[0087] This embodiment provides a low-light image enhancement system, such as Figure 6 As shown, it includes an acquisition unit, a processing unit, a first calculation unit, and a second calculation unit, wherein:
[0088] The acquisition unit is used to acquire low-light image data of the target.
[0089] The processing unit is used to import the target low-light image data into a preset image enhancement network model for image enhancement processing to obtain weight parameters, global enhancement map data, and local adjustment map data. The image enhancement network model includes an image preprocessing network, a first processing channel, a second processing channel, and a third processing channel. The image preprocessing network is used to preprocess the target low-light image data to extract several feature map data. The first processing channel is used to perform image regression processing on the several feature map data to obtain weight parameters. The second processing channel is used to perform global enhancement processing on the several feature map data to obtain global enhancement map data. The third processing channel is used to perform local adjustment processing on the several feature map data to obtain local adjustment map data.
[0090] The first calculation unit is used to calculate the inverse illumination map data based on the weight parameters, global enhancement map data, and local adjustment map data;
[0091] The second calculation unit is used to perform a Hadamard product operation on the inverse illumination map data and the target low-light image data to obtain the enhanced image data of the target low-light image.
[0092] Example 3:
[0093] This embodiment provides a low-light image enhancement device, such as... Figure 7 As shown, at the hardware level, it includes:
[0094] The data interface is used to establish data communication between the processor and external devices;
[0095] Memory, used to store instructions;
[0096] The processor is configured to read instructions stored in the memory and execute the low-light image enhancement method in Embodiment 1 according to the instructions.
[0097] Optionally, the computer device also includes an internal bus. The processor, memory, and data interface can be interconnected via the internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc.
[0098] The memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory. The processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0099] Example 4:
[0100] This embodiment provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the low-light image enhancement method of Embodiment 1. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems.
[0101] This embodiment also provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the low-light image enhancement method of Embodiment 1. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system.
[0102] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A low-light image enhancement method, characterized in that, include: Acquire low-light image data of the target; The target low-light image data is imported into a pre-set image enhancement network model for image enhancement processing to obtain weight parameters, global enhancement map data, and local adjustment map data. The image enhancement network model includes an image preprocessing network, a first processing channel, a second processing channel, and a third processing channel. The image preprocessing network is used to preprocess the target low-light image data to extract several feature map data. The first processing channel is used to perform image regression processing on the several feature map data to obtain weight parameters. The second processing channel is used to perform global enhancement processing on the several feature map data to obtain global enhancement map data. The third processing channel is used to perform local adjustment processing on the several feature map data to obtain local adjustment map data. Inverse illumination map data is calculated based on weight parameters, global enhancement map data, and local adjustment map data; The inverse illumination map data and the target low-light image data are subjected to a Hadamard product operation to obtain the target low-light image enhanced image data.
2. The low-light image enhancement method according to claim 1, characterized in that, The preprocessing network uses the ResNet50 model to preprocess the target low-light image data to extract several feature map data.
3. The low-light image enhancement method according to claim 1, characterized in that, The first processing channel includes a regression module, which includes a convolutional kernel, an average pooling kernel, a multilayer perceptron, and a sigmoid layer connected in sequence. The sigmoid layer outputs weight parameters in the range (0,1).
4. The low-light image enhancement method according to claim 1, characterized in that, The second processing channel includes a straightening module, a position encoding module, a SwinTransformer module, and a first TaskTail Block module connected in sequence. The straightening module is used to straighten the feature map data. The position encoding module is used to perform position encoding on the output of the straightening module to obtain a first tokens vector. The SwinTransformer module is used to input the first tokens vector to perform global modeling and multi-scale feature extraction, and output a second tokens vector. The first TaskTail Block module is used to input the second tokens vector to perform linear mapping and rearrangement processing, and output global augmented map data.
5. The low-light image enhancement method according to claim 1, characterized in that, The third processing channel includes a second Task Tail Block module, which is used to input several feature map data for linear mapping and rearrangement processing, and output local adjustment map data.
6. The low-light image enhancement method according to claim 1, characterized in that, The loss function used in the image enhancement network model is: L total =L spa +λ1L exp +λ2L col +λ3L smo Among them, L total Let L be the overall loss function of the image enhancement network model. spa Characteristic space consistency loss function, L exp Characterizing the exposure control loss function, L col The color consistency loss function, L smo The loss function characterizing illuminance smoothness, λ1, λ2, and λ3 are respectively L exp L col and L smo The weight.
7. The low-light image enhancement method according to claim 1, characterized in that, The formula for calculating the inverse illumination map data based on the weight parameters, global enhancement map data, and local adjustment map data is as follows: in, For inverse illumination map data, For globally enhanced graph data, For locally adjusted plot data, α is the weighting parameter; The formula for calculating the Hadamard product of the inverse illumination map data and the target low-light image data is as follows: Where R represents the enhanced low-light image data of the target, and L represents the low-light image data of the target. Characterizes the Hadamard product operator.
8. A low-light image enhancement system, characterized in that, It includes an acquisition unit, a processing unit, a first calculation unit, and a second calculation unit, wherein: The acquisition unit is used to acquire low-light image data of the target. The processing unit is used to import the target low-light image data into a preset image enhancement network model for image enhancement processing to obtain weight parameters, global enhancement map data, and local adjustment map data. The image enhancement network model includes an image preprocessing network, a first processing channel, a second processing channel, and a third processing channel. The image preprocessing network is used to preprocess the target low-light image data to extract several feature map data. The first processing channel is used to perform image regression processing on the several feature map data to obtain weight parameters. The second processing channel is used to perform global enhancement processing on the several feature map data to obtain global enhancement map data. The third processing channel is used to perform local adjustment processing on the several feature map data to obtain local adjustment map data. The first calculation unit is used to calculate the inverse illumination map data based on the weight parameters, global enhancement map data, and local adjustment map data; The second calculation unit is used to perform a Hadamard product operation on the inverse illumination map data and the target low-light image data to obtain the enhanced image data of the target low-light image.
9. A low-light image enhancement device, characterized in that, include: Memory, used to store instructions; A processor is configured to read instructions stored in the memory and execute the method according to any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the method described in any one of claims 1-7.