A method of adaptive regional low-light enhancement and related devices

Abstract

The application discloses a method for self-adaptive region low-light enhancement and related equipment, and the method comprises the following steps: acquiring a low-illumination image, inputting the low-illumination image into a perception guide map module to obtain guide image information; obtaining a guide map based on the guide image information, inputting the guide map into a multi-scale dilation convolution attention mechanism module to obtain a first image after detail repair; performing a global average pooling operation on the first image to obtain a second image after enhancement, and processing the second image to obtain a low-light enhancement image. The purpose of the application is to improve the visual effect under low-light conditions, to make the neural network adaptively enhance the low-light image in a region-dependent manner by proposing a self-adaptive learning guide map, and to further assimilate the perception information of the adaptive guide map through a multi-scale receptive field module, so that the model learns more abundant and more real texture details.

Description

Technical Field

[0001] This invention relates to the field of image enhancement technology, and in particular to a method, system, terminal, and computer-readable storage medium for adaptive regional low-light enhancement. Background Technology

[0002] In the process of building smart cities, image recognition, video surveillance, and traffic management systems have been widely applied. However, in low-light nighttime scenarios, the captured images and videos suffer from reduced visual quality or contrast due to ambient noise or noise from the acquisition equipment. For example, regions of interest may be difficult to identify, and smooth areas may contain significant noise or lack of detail. Low-light enhancement aims to recover normally lit images from low-quality images under insufficient lighting conditions, thereby improving the overall image quality. This technology has important applications in both civilian video surveillance systems and military reconnaissance systems.

[0003] Current methods primarily employ adjustment-based image enhancement, such as Retinex enhancement. Adjustment-based low-light image enhancement schemes can be categorized into global histogram equalization (HMA) and local histogram equalization (MHE). Histogram equalization stretches the dynamic range of low-light images. Global histogram equalization adjusts the brightness of the entire input image, ignoring local contextual consistency and potentially introducing overexposure and amplified noise. Conversely, local histogram equalization requires dividing the image into sub-blocks for local adjustments, resulting in high computational cost and compromising real-time performance. Retinex-based image enhancement algorithms essentially decompose the input image into illumination and reflection layers, with accurate estimation being crucial. However, Retinex-based algorithms do not consider gradient information, leading to image blurring during enhancement. Consequently, existing low-light enhancement algorithms lack the ability to enhance different regions within low-light images, thus failing to effectively adaptively enhance images with varying backgrounds.

[0004] Therefore, existing technologies still need to be improved and developed. Summary of the Invention

[0005] The main objective of this invention is to provide an adaptive low-light enhancement method and related equipment, which aims to solve the problem that existing low-light enhancement algorithms cannot enhance different regions in low-light images, thus failing to effectively adaptively enhance images with different backgrounds.

[0006] To achieve the above objectives, the present invention provides a method and related equipment for adaptive region low-light enhancement, wherein the method for adaptive region low-light enhancement includes the following steps:

[0007] A low-light image is acquired and input into the perception guidance map module to obtain guidance image information;

[0008] Based on the guidance image information, a guidance map is obtained. The guidance map is then input into the multi-scale dilated convolutional attention mechanism module to obtain the first image after detail restoration.

[0009] A global average pooling operation is performed on the first image to obtain an enhanced second image, and the second image is further processed to obtain a low-light enhanced image.

[0010] Optionally, in the adaptive region low-light enhancement method, the perception guidance map module consists of five residual blocks, one long short-term memory unit, and one convolutional layer.

[0011] Optionally, the adaptive region low-light enhancement method, wherein obtaining a guide map based on the guide image information and inputting the guide map into a multi-scale dilated convolutional attention mechanism module to obtain a first image with detail restoration specifically includes:

[0012] Different input scenarios are acquired, and a corresponding guide image is generated based on the guide image information.

[0013] The guide image is input to the multi-scale dilated convolutional attention mechanism module, which repairs the image details of different regions of the guide image using feature information of different granularities, to obtain the first image after detail repair.

[0014] Optionally, the adaptive region low-light enhancement method, wherein after acquiring different input scenes and generating a guide map corresponding to the input scene based on the guide image information, further includes:

[0015] The guide map is subjected to smoothing constraint processing to repair the noise in the guide map, thereby obtaining the target guide map.

[0016] Optionally, in the adaptive region low-light enhancement method, the multi-scale dilated convolutional attention mechanism module consists of three cascaded multi-scale dilated convolutional attention mechanism structures, wherein each multi-scale dilated convolutional attention mechanism structure consists of three parallel features, and each of the parallel features contains a different receptive field.

[0017] Optionally, the adaptive region low-light enhancement method, wherein performing a global average pooling operation on the first image to obtain an enhanced second image, and processing the second image to obtain a low-light enhanced image, specifically includes:

[0018] The enhanced features are obtained, and a global average pooling operation is performed on the first image based on the enhanced features to obtain the enhanced second image;

[0019] The second image is constrained by the perceptual loss of the overall loss function to obtain a low-light enhanced image.

[0020] Optionally, in the adaptive regional low-light enhancement method, the overall loss function is:

[0021] L Total =λ*L TV +L Percept +L Fid ;

[0022] Among them, L Total Let L be the overall loss function, where λ is a constant. TV For the total variation, L Percept To perceive loss, L Fid For constraints.

[0023] Optionally, in the adaptive region low-light enhancement method, the enhancement feature is composed of a three-branch feature aggregation; wherein the expression of the enhancement feature is:

[0024]

[0025] in, To enhance features, The first characteristic, This is the second characteristic. This is the third characteristic.

[0026] Optionally, in the adaptive region low-light enhancement method, the first feature is: The second characteristic is: The third characteristic is:

[0027] Where s represents weight learning, U1 represents the first parallel feature, U2 represents the second parallel feature, and U3 represents the third parallel feature.

[0028] Optionally, in the adaptive region low-light enhancement method, the preset factors include image brightness, contrast, and structure.

[0029] Furthermore, to achieve the above objectives, the present invention also provides an adaptive regional low-light enhancement system, wherein the adaptive regional low-light enhancement system comprises:

[0030] The data acquisition module is used to acquire low-light images and input the low-light images into the perception guidance map module to obtain guidance image information;

[0031] The image processing module is used to obtain a guide map based on the guide image information, and input the guide map into the multi-scale dilated convolutional attention mechanism module to obtain the first image after detail restoration;

[0032] The result generation module is used to perform global average pooling on the first image to obtain an enhanced second image, and to process the second image to obtain a low-light enhanced image.

[0033] In addition, to achieve the above objectives, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and an adaptive region low light enhancement program stored in the memory and executable on the processor, wherein when the adaptive region low light enhancement program is executed by the processor, it implements the steps of the adaptive region low light enhancement method as described above.

[0034] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an adaptive region low-light enhancement program, which, when executed by a processor, implements the steps of the adaptive region low-light enhancement method as described above.

[0035] In this invention, a low-light image is acquired and input into a perceptual guidance map module to obtain guidance image information. Based on this guidance image information, a guidance map is obtained and input into a multi-scale dilated convolutional attention mechanism module to obtain a first image with detail restoration. Global average pooling is performed on the first image to obtain an enhanced second image, which is then processed to obtain a low-light enhanced image. The purpose of this invention is to improve visual effects under low-light conditions by proposing an adaptive learning guidance map, enabling the neural network to adaptively enhance low-light images in a region-dependent manner. Furthermore, a multi-scale receptive field module further assimilates the perceptual information from the adaptive guidance map, allowing the model to learn richer and more realistic texture details. Attached Figure Description

[0036] Figure 1 This is a flowchart of a preferred embodiment of the adaptive region low-light enhancement method of the present invention;

[0037] Figure 2 This is a flowchart of step S20 in a preferred embodiment of the adaptive region low-light enhancement method of the present invention;

[0038] Figure 3 This is a flowchart of step S30 in a preferred embodiment of the adaptive region low-light enhancement method of the present invention;

[0039] Figure 4This is a schematic diagram of the overall process of the adaptive region low-light enhancement method in this invention;

[0040] Figure 5 This is a detailed schematic diagram of the multi-scale dilated convolutional attention mechanism module in the adaptive region low-light enhancement method of this invention;

[0041] Figure 6 This is a schematic diagram of a preferred embodiment of the adaptive region low-light enhancement system of the present invention;

[0042] Figure 7 This is a schematic diagram of the operating environment of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0044] It should be noted that if the embodiments of the present invention involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of the components in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0045] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0046] The adaptive region low-light enhancement method described in the preferred embodiment of the present invention, such as... Figure 1 As shown, the adaptive region low-light enhancement method includes the following steps:

[0047] Step S10: Obtain a low-light image and input the low-light image into the perception guidance map module to obtain guidance image information.

[0048] Specifically, a low-light image is acquired and input into a perceptual guidance map module. This module consists of five residual blocks, one long short-term memory (LSTM) unit, and one convolutional layer, used to generate rich guidance image information (denoted by G). Different input scenes will adaptively generate corresponding guidance maps, which serve as additional information to aid image recovery. By adaptively generating guidance maps that enhance different regions, detail compensation prediction for the low-light image is achieved. To avoid unavoidable noise in the generated guidance map, smoothing constraints are applied during image enhancement to correct it without affecting the quality of the recovered image. The process of inputting the low-light image (denoted by I) can be represented by the following formula: G = F(I), where G is the guidance image information, I is the low-light image, and F(I) is the function for processing the low-light image.

[0049] Step S20: Obtain a guide map based on the guide image information, and input the guide map into the multi-scale dilated convolutional attention mechanism module to obtain the first image after detail restoration.

[0050] Please refer to the detailed process. Figure 2 This is a flowchart of step S20 in the adaptive region low-light enhancement method provided by the present invention.

[0051] like Figure 2 As shown, step S20 includes:

[0052] Step S21: Obtain different input scenarios and generate a guide image corresponding to the input scenario based on the guide image information;

[0053] Step S22: Input the guide map into the multi-scale dilated convolutional attention mechanism module. The multi-scale dilated attention mechanism module repairs the image details of different regions of the guide map using feature information of different granularities to obtain the first image after detail repair.

[0054] Specifically, different input scenarios are acquired, and a guidance map corresponding to the input scenario is generated based on the guidance image information. The guidance map is then input to the multi-scale dilated convolutional attention mechanism module. The core of the multi-scale dilated convolutional attention mechanism module consists of three cascaded MSDCA (Multi-scale Dilated Convolution Attention) structures. Each MSDCA structure consists of three parallel features (the first parallel feature, the second parallel feature, and the third parallel feature are represented by U1, U2, and U3, respectively). Each parallel feature has a different receptive field. In this way, feature information of different granularities is used to recover image details in different regions, resulting in a first image after detail restoration. By using a module that combines multi-scale dilated convolution with the attention mechanism, the model learns information of different regions and different granularities, and assimilates and absorbs the perceptual information of the guidance map and the multi-scale convolutional information in an end-to-end manner, thereby enhancing the texture details in different regions.

[0055] Step S30: Perform global average pooling on the first image to obtain an enhanced second image, and process the second image to obtain a low-light enhanced image.

[0056] Please refer to the detailed process. Figure 3 This is a flowchart of step S30 in the adaptive region low-light enhancement method provided by the present invention.

[0057] like Figure 3 As shown, step S30 includes:

[0058] Step S31: Obtain the enhancement features, and perform global average pooling on the first image based on the enhancement features to obtain the enhanced second image;

[0059] Step S32: Constrain the preset factors of the second image according to the perceptual loss of the overall loss function to obtain a low-light enhanced image.

[0060] Specifically, to learn weights for different channels (denoted by s), a learnable attention mechanism enhances important regions of features by performing global average pooling and softmax on the first image, resulting in an enhanced second image; wherein, the enhanced features used... It is composed of these three features (the first feature, the second feature, and the third feature, respectively) and (represented by) aggregation; the expression for the enhanced feature is: in, To enhance features, The first characteristic, This is the second characteristic. The third feature is used; the second image is processed according to the overall loss function to obtain a low-light enhanced image; wherein, the overall loss function consists of the following three parts:

[0061] L Total =*L TV +L Percept +L Fid ;

[0062] Among them, L Total Let L be the overall loss function, where λ is a constant. TV For the total variation, L Percept To perceive loss, L Fid For constraints; in the total variation (L TV ( ) is used in image restoration tasks to eliminate ghosting and noise in guide maps; while using perceptual loss (L Percept The enhanced image is compared with the target image at the feature layer, making the output image very close to the target image. Furthermore, structural similarity (SSIM) loss is also used as a constraint (L). Fid The target image is approximated by three different factors: image brightness, contrast, and structure; λ is a constant, and here λ is set to 0.0001. The technical solution proposed in this invention can enhance low-light images or videos and improve the visual effect under low-light conditions. In addition, this invention uses two commonly used full-reference quality evaluation indicators, structural similarity (SSIM)[1] and peak signal-to-noise ratio (PSNR), to measure and compare the performance of the enhancement algorithm. The larger the values ​​of these two indicators, the better the performance of the algorithm. This invention can show competitive performance in quantitative results. For specific results, please refer to the results of the test dataset in the LOL[2] dataset in the following table.

[0063]

[0064] Furthermore, such as Figure 4 As shown, the overall process of this invention specifically includes: acquiring a low-light image; inputting the low-light image into a perceptual guidance map module to obtain guidance image information; obtaining a guidance map based on the guidance image information; concatenating the low-light image and the guidance map together (C in the figure) and inputting them forward into a multi-scale dilated convolutional attention mechanism module to obtain a first image with detail restoration; performing global average pooling on the first image to obtain an enhanced second image; and processing the second image to obtain a low-light enhanced image. A detailed diagram of the multi-scale dilated convolutional attention mechanism module is shown below. Figure 5As shown in the figure, U represents the total parallelism feature, which includes three different parallelism features U1, U2, and U3 (for example, U1 represents the first parallelism feature, U2 represents the second parallelism feature, and U3 represents the third parallelism feature). The enhanced features are derived from these three features (the first feature, the second feature, and the third feature, respectively) using... and GAP / S represents global average pooling and softmax operations.

[0065] Furthermore, such as Figure 6 As shown, based on the above-described adaptive region low-light enhancement method, the present invention also provides a corresponding adaptive region low-light enhancement system, wherein the adaptive region low-light enhancement system includes:

[0066] Data acquisition module 51 is used to acquire low-light images and input the low-light images into the perception guidance map module to obtain guidance image information;

[0067] Image processing module 52 is used to obtain a guide map based on the guide image information, and input the guide map into the multi-scale dilated convolutional attention mechanism module to obtain the first image after detail restoration;

[0068] The result generation module 53 is used to perform a global average pooling operation on the first image to obtain an enhanced second image, and to process the second image to obtain a low-light enhanced image.

[0069] Furthermore, such as Figure 7 As shown, based on the above-described adaptive regional low-light enhancement method and system, the present invention also provides a terminal, which includes a processor 10, a memory 20, and a display 30. Figure 7 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0070] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a panoramic video storage optimization program 40, which can be executed by the processor 10 to implement the adaptive low-light enhancement method of this application.

[0071] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as performing the adaptive region low-light enhancement method.

[0072] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other via a system bus.

[0073] In one embodiment, when the processor 10 executes the storage optimization program 40 for the panoramic video in the memory 20, the following steps are performed:

[0074] A low-light image is acquired and input into the perception guidance map module to obtain guidance image information;

[0075] Based on the guidance image information, a guidance map is obtained. The guidance map is then input into the multi-scale dilated convolutional attention mechanism module to obtain the first image after detail restoration.

[0076] A global average pooling operation is performed on the first image to obtain an enhanced second image, and the second image is further processed to obtain a low-light enhanced image.

[0077] The perception guidance graph module consists of five residual blocks, one long short-term memory unit, and one convolutional layer.

[0078] Specifically, obtaining a guidance map based on the guidance image information and inputting the guidance map into a multi-scale dilated convolutional attention mechanism module to obtain a first image after detail restoration includes:

[0079] Different input scenarios are acquired, and a corresponding guide image is generated based on the guide image information.

[0080] The guide image is input to the multi-scale dilated convolutional attention mechanism module, which repairs the image details of different regions of the guide image using feature information of different granularities, to obtain the first image after detail repair.

[0081] The step of acquiring different input scenarios and generating a guidance map corresponding to the input scenario based on the guidance image information further includes:

[0082] The guide map is subjected to smoothing constraint processing to repair the noise in the guide map, thereby obtaining the target guide map.

[0083] The multi-scale dilated convolutional attention mechanism module consists of three cascaded multi-scale dilated convolutional attention mechanism structures, each of which consists of three parallel features, and each of the parallel features contains a different receptive field.

[0084] Specifically, the step of performing a global average pooling operation on the first image to obtain an enhanced second image, and then processing the second image to obtain a low-light enhanced image, includes:

[0085] The enhanced features are obtained, and a global average pooling operation is performed on the first image based on the enhanced features to obtain the enhanced second image;

[0086] The second image is constrained by the perceptual loss of the overall loss function to obtain a low-light enhanced image.

[0087] The overall loss function is:

[0088] L Total =λ*L TV +L Percept +L Fid ;

[0089] Among them, L Total Let L be the overall loss function, where λ is a constant. TV For the total variation, L Percept To perceive loss, L FidFor constraints.

[0090] The enhancement feature is composed of three feature aggregations; the expression for the enhancement feature is:

[0091]

[0092] in, To enhance features, The first characteristic, This is the second characteristic. This is the third characteristic.

[0093] The first feature is: The second characteristic is: The third characteristic is:

[0094] Where s represents weight learning, U1 represents the first parallel feature, U2 represents the second parallel feature, and U3 represents the third parallel feature.

[0095] The preset factors include image brightness, contrast, and structure.

[0096] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an adaptive region low-light enhancement program, which, when executed by a processor, implements the steps of the adaptive region low-light enhancement method as described above.

[0097] In summary, this invention provides an adaptive low-light enhancement method, comprising: acquiring a low-light image; inputting the low-light image into a perceptual guidance map module to obtain guidance image information; obtaining a guidance map based on the guidance image information; inputting the guidance map into a multi-scale dilated convolutional attention mechanism module to obtain a first image with detail restoration; performing global average pooling on the first image to obtain an enhanced second image; and processing the second image to obtain a low-light enhanced image. The purpose of this invention is to improve visual effects under low-light conditions by proposing an adaptive learning guidance map, enabling the neural network to adaptively enhance low-light images in a region-dependent manner, and further assimilating the perceptual information of the adaptive guidance map through a multi-scale receptive field module, thereby allowing the model to learn richer and more realistic texture details.

[0098] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0099] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The computer-readable storage medium can be a memory, magnetic disk, optical disk, etc.

[0100] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for adaptive low-light enhancement in a specific region, characterized in that, The adaptive region low-light enhancement method includes: A low-light image is acquired and input into the perception guidance map module to obtain guidance image information; Based on the guidance image information, a guidance map is obtained. The guidance map is then input into the multi-scale dilated convolutional attention mechanism module to obtain the first image after detail restoration. The process of obtaining a guidance map based on the guidance image information, and inputting the guidance map into a multi-scale dilated convolutional attention mechanism module to obtain a first image after detail restoration, specifically includes: Different input scenarios are acquired, and a corresponding guide image is generated based on the guide image information. The guide image is input to the multi-scale dilated convolutional attention mechanism module, which repairs the image details of different regions of the guide image using feature information of different granularities, to obtain the first image after detail repair. A global average pooling operation is performed on the first image to obtain an enhanced second image, and the second image is further processed to obtain a low-light enhanced image; The process of performing global average pooling on the first image to obtain an enhanced second image, and then processing the second image to obtain a low-light enhanced image, specifically includes: The enhanced features are obtained, and a global average pooling operation is performed on the first image based on the enhanced features to obtain the enhanced second image; The second image is constrained by the perceptual loss of the overall loss function to obtain a low-light enhanced image, thereby enhancing the low-light image and improving the visual effect under low-light conditions.

2. The adaptive region low-light enhancement method according to claim 1, characterized in that, The perception guidance graph module consists of five residual blocks, one long short-term memory unit, and one convolutional layer.

3. The adaptive regional low-light enhancement method according to claim 1, characterized in that, The process of acquiring different input scenarios and generating a guidance map corresponding to the input scenario based on the guidance image information further includes: The guide map is subjected to smoothing constraint processing to repair the noise in the guide map, thereby obtaining the target guide map.

4. The adaptive region low-light enhancement method according to claim 1, characterized in that, The multi-scale dilated convolutional attention mechanism module consists of three cascaded multi-scale dilated convolutional attention mechanism structures, wherein each multi-scale dilated convolutional attention mechanism structure consists of three parallel features, and each parallel feature contains a different receptive field.

5. The adaptive region low-light enhancement method according to claim 1, characterized in that, The overall loss function is: + + ； in, For the overall loss function, It is a constant. For the total variation, In order to perceive loss, For constraints.

6. The adaptive region low-light enhancement method according to claim 5, characterized in that, The enhancement feature is composed of three feature aggregations; wherein, the expression of the enhancement feature is: + + ； in, To enhance features, The first characteristic, This is the second characteristic. This is the third characteristic.

7. The adaptive region low-light enhancement method according to claim 6, characterized in that, The first feature is: The second characteristic is: The third characteristic is: ; in, For weight learning, This is the first parallel feature. This is the second parallel feature. This is the third parallel feature.

8. The method for adaptive regional low-light enhancement according to claim 1, characterized in that, The preset factors include image brightness, contrast, and structure.

9. A system for adaptive regional low-light enhancement, characterized in that, The adaptive region low-light enhancement system is used to implement the adaptive region low-light enhancement method according to any one of claims 1-8, wherein the adaptive region low-light enhancement system comprises: The data acquisition module is used to acquire low-light images and input the low-light images into the perception guidance map module to obtain guidance image information; The image processing module is used to obtain a guide map based on the guide image information, and input the guide map into the multi-scale dilated convolutional attention mechanism module to obtain the first image after detail restoration; The result generation module is used to perform global average pooling on the first image to obtain an enhanced second image, and to process the second image to obtain a low-light enhanced image.

10. A terminal, characterized in that, The terminal includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the adaptive regional low-light enhancement method as described in any one of claims 1-8.

11. A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of the adaptive regional low-light enhancement method as claimed in any one of claims 1-8.

Images