Low-light image enhancement methods, apparatuses, electronic devices, and computer-readable media

Abstract

This disclosure provides embodiments of a low-light image enhancement method, apparatus, electronic device, and computer-readable medium. One specific implementation of the method includes: acquiring an initial low-light image; performing image quality detection processing on the initial low-light image; in response to determining that the image quality detection information does not meet preset detection conditions, sending exposure adjustment information corresponding to the image quality detection information to a low-light camera device to adjust the sensitivity and exposure time of the low-light camera device; and acquiring the low-light image again using the adjusted low-light camera device; obtaining a normal-light reference image; inputting the normal-light reference image into a reference image feature extraction cooperative network of a pre-trained low-light image enhancement model; and inputting the low-light image and reference image feature information into an image enhancement association network to obtain an enhanced low-light image corresponding to the low-light image. This implementation improves the image quality and image sharpness of the enhanced low-light image.

Description

Technical Field

[0001] Embodiments of this disclosure relate to the field of computer technology, and more particularly to low-light image enhancement methods, apparatus, electronic devices, and computer-readable media. Background Technology

[0002] Low-light image enhancement is a technique for enhancing low-light images captured by low-light cameras (e.g., low-light cameras). Currently, the common method for enhancing low-light images captured by low-light cameras is to directly enhance the low-light images captured by the low-light camera.

[0003] However, when using the above methods to enhance low-light images captured by low-light cameras, the following technical problems often arise:

[0004] Directly enhancing low-light images captured by low-light cameras without optimizing the acquisition process based on the low-light environment means the original image may contain significant noise. This noise is amplified during enhancement, potentially making the enhanced low-light image more grainy and coarser. Furthermore, while directly enhancing low-light images can improve brightness and contrast to some extent, the poor quality of the original image limits the overall quality of the enhanced image, resulting in a still somewhat blurry image.

[0005] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not form prior art known to those skilled in the art. Summary of the Invention

[0006] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0007] Some embodiments of this disclosure provide low-light image enhancement methods, apparatuses, electronic devices, and computer-readable media to address one or more of the technical problems mentioned in the background section above.

[0008] In a first aspect, some embodiments of this disclosure provide a low-light image enhancement method, the method comprising: acquiring an initial low-light image using a low-light camera; performing image quality detection processing on the initial low-light image to obtain image quality detection information; determining the initial low-light image as a low-light image in response to determining that the image quality detection information meets preset detection conditions; sending exposure adjustment information corresponding to the image quality detection information to the low-light camera in response to determining that the image quality detection information does not meet preset detection conditions, thereby adjusting the sensitivity and exposure time of the low-light camera, and acquiring the low-light image again using the adjusted low-light camera; acquiring a normal light reference image; inputting the normal light reference image into a reference image feature extraction collaborative network of a pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal light reference image, wherein the low-light image enhancement model includes the reference image feature extraction collaborative network and an image enhancement association network; and inputting the low-light image and the reference image feature information into the image enhancement association network to obtain an enhanced low-light image corresponding to the low-light image.

[0009] Secondly, some embodiments of this disclosure provide a low-light image enhancement device, comprising: a first acquisition unit configured to acquire an initial low-light image via a low-light camera; an image quality detection unit configured to perform image quality detection processing on the initial low-light image to obtain image quality detection information; a determination unit configured to determine the initial low-light image as a low-light image in response to determining that the image quality detection information meets preset detection conditions; and a second acquisition unit configured to send exposure adjustment information corresponding to the image quality detection information to the low-light camera in response to determining that the image quality detection information does not meet preset detection conditions, so as to adjust the low-light camera. The device's photosensitivity and exposure time, and the low-light image acquired again by the adjusted low-light camera device; an acquisition unit is configured to acquire a normal light reference image; a first input unit is configured to input the normal light reference image into a pre-trained low-light image enhancement model's reference image feature extraction collaborative network to obtain reference image feature information corresponding to the normal light reference image, wherein the low-light image enhancement model includes the reference image feature extraction collaborative network and an image enhancement association network; a second input unit is configured to input the low-light image and the reference image feature information into the image enhancement association network to obtain an enhanced low-light image corresponding to the low-light image.

[0010] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0011] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0012] The various embodiments of this disclosure have the following beneficial effects: the low-light image enhancement methods of some embodiments of this disclosure improve the image quality and image clarity of the enhanced low-light image. Specifically, the reason for the poor image quality and image clarity of the enhanced low-light image is that: directly enhancing the low-light image captured by the low-light camera device does not optimize the acquisition process based on the low-light image captured in the low-light environment. The original image captured without optimization may contain a lot of noise, which is also amplified during the enhancement process, making the noise in the enhanced low-light image more obvious, resulting in a more grainy and coarser image quality. At the same time, while directly enhancing the low-light image captured by the low-light camera device can improve the brightness and contrast of the low-light image to some extent, the quality of the enhanced image may still be limited due to the poor quality of the original image, resulting in a relatively blurry enhanced low-light image. Based on this, the low-light image enhancement method of some embodiments of this disclosure first acquires an initial low-light image using a low-light camera device. Thus, an initial low-light image can be obtained to determine whether to re-acquire the low-light image. Then, image quality detection processing is performed on the initial low-light image to obtain image quality detection information. This allows for image quality detection of the initial low-light image to generate image quality detection information. This image quality detection information characterizes the image quality acquired by the low-light camera device in a low-light environment without optimized acquisition process. Subsequently, in response to determining that the image quality detection information meets preset detection conditions, the initial low-light image is identified as a low-light image. Thus, given that the image quality detection information of the initial low-light image meets the preset detection conditions, an initial low-light image with relatively good image quality can be identified as a low-light image. Then, in response to determining that the image quality detection information does not meet the preset detection conditions, exposure adjustment information corresponding to the image quality detection information is sent to the low-light camera device to adjust the sensitivity and exposure time of the low-light camera device, and the low-light image is acquired again using the adjusted low-light camera device. Thus, even when the image quality of the initial low-light image is poor (i.e., the image quality detection information does not meet the preset detection conditions), the sensitivity and exposure time of the low-light camera device can be adjusted to optimize the parameters of the low-light camera device and improve the acquisition process. After quality inspection, adjusting the ISO and exposure time allows the low-light camera to better adapt to low-light environments, capturing relatively high-quality low-light images and reducing noise introduction. Then, a normal-light reference image is acquired. This provides a normal-light reference image for generating reference image feature information. Next, the normal-light reference image is input into the reference image feature extraction collaborative network of a pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal-light reference image. The low-light image enhancement model includes the reference image feature extraction collaborative network and an image enhancement association network.Therefore, reference image feature information of a normal light reference image can be extracted through a reference image feature extraction collaborative network. Then, the aforementioned low-light image and the aforementioned reference image feature information are input into the aforementioned image enhancement association network to obtain an enhanced low-light image corresponding to the aforementioned low-light image. Thus, the low-light image can be enhanced based on the reference image feature information through the image enhancement association network to obtain an enhanced low-light image. Furthermore, because the initial low-light image acquired by the low-light camera device was quality-checked before enhancement, and the sensitivity and exposure time of the low-light camera device were adjusted to optimize the acquisition process, the parameters of the low-light camera device were optimized. Adjusting the sensitivity and exposure time after quality check allows the low-light camera device to better adapt to low-light environments, acquire relatively high-quality low-light images, reduce the introduction of noise in the low-light image itself, and improve the image quality and sharpness of the enhanced low-light image. Attached Figure Description

[0013] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0014] Figure 1 This is a flowchart of some embodiments of the low-light image enhancement method according to the present disclosure;

[0015] Figure 2 This is a schematic diagram of the structure of some embodiments of the low-light image enhancement apparatus according to the present disclosure;

[0016] Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure;

[0017] Figure 4 This is a schematic diagram of a model structure based on some embodiments of the low-light image enhancement model of this disclosure;

[0018] Figure 5 This is another schematic diagram of a model structure based on some embodiments of the low-light image enhancement model of this disclosure;

[0019] Figure 6 This is a schematic diagram of the network structure according to some embodiments of the image enhancement association network of this disclosure;

[0020] Figure 7 This is a schematic diagram of the network structure of some embodiments of the reference image feature extraction cooperative network according to the present disclosure. Detailed Implementation

[0021] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0022] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0023] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0024] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0025] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0026] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0027] Figure 1 A flow 100 of some embodiments of a low-light image enhancement method according to the present disclosure is shown. The low-light image enhancement method includes the following steps:

[0028] Step 101: Acquire an initial low-light image using a low-light camera device.

[0029] In some embodiments, the entity executing the low-light image enhancement method (e.g., a computing device) can acquire an initial low-light image using a low-light camera. This low-light camera can be an image acquisition device (e.g., a low-light camera) that acquires images in low-light environments (e.g., at night or in dimly lit environments). The initial low-light image can be an image captured in a low-light environment without adjusting the sensitivity and exposure time of the low-light camera.

[0030] Step 102: Perform image quality detection processing on the initial low-light image to obtain image quality detection information.

[0031] In some embodiments, the execution entity may perform image quality detection processing on the initial low-light image to obtain image quality detection information.

[0032] In the process of adopting technical solutions to address the problems mentioned in the background section, the following issues often arise:

[0033] When performing quality inspection on low-light images, the color features of RGB low-light images have complex correlations. Changes in the values ​​of the red (R), green (G), and blue (B) channels can simultaneously affect the type, intensity, and brightness of colors. Directly performing quality inspection on RGB low-light images through an image quality assessment model increases the complexity of the model. The model requires more computing resources to perform quality inspection on RGB low-light images, which increases the waste of computing resources during the quality inspection process.

[0034] In response to the aforementioned technical problems, the following solution was adopted:

[0035] In some optional implementations of certain embodiments, the aforementioned execution entity may perform image quality detection processing on the initial low-light image through the following steps to obtain image quality detection information:

[0036] The first step is to determine the area of ​​the initial low-light image as the low-light image partition area.

[0037] The second step is to divide the initial low-light image into blocks with an area equal to the preset division area in response to the determination that the area of ​​the low-light image division is divisible by the preset division area.

[0038] Thirdly, in response to the determination that the area of ​​the low-light image segmentation cannot be divided by the preset segmentation area, the initial low-light image is segmented using a sliding window based on a preset sliding window length, preset sliding window width, preset horizontal sliding step size, and preset vertical sliding step size to obtain individual initial low-light image blocks. In practice, the execution entity can execute a pre-encapsulated segmentation method to segment the initial low-light image based on the aforementioned sliding window length, sliding window width, horizontal sliding step size, and vertical sliding step size to obtain individual initial low-light image blocks. There are overlapping areas among the individual initial low-light image blocks segmented by the sliding window.

[0039] Fourth, for each initial low-light image patch obtained from the initial low-light image patches, perform the following steps:

[0040] The first sub-step involves performing the following steps for each pixel in the initial low-light image block:

[0041] Sub-step one involves determining the pixel value information of the initial low-light image block pixels as the pixel value information to be converted. This pixel value information can be the Rgb data of the initial low-light image block pixels. The pixel value information can include red pixel values ​​(R), green pixel values ​​(G), and blue pixel values ​​(B). For example, the pixel value information could be "(R=255, G=0, B=0)".

[0042] Sub-step two involves normalizing the information to be converted to obtain normalized pixel value information. In practice, the executing entity can determine the ratio of the red pixel value to a preset value in the information to be converted as the updated red pixel value, the ratio of the green pixel value to the preset value in the information to be converted as the updated green pixel value, and the ratio of the blue pixel value to the preset value as the updated blue pixel value. Then, the executing entity can determine the updated red pixel value, the updated green pixel value, and the updated blue pixel value as the normalized pixel value information. The preset value can be 255. As an example, the information to be converted can be "(R=255, G=0, B=0)", and the normalized pixel value information can be "(R=1, G=0, B=0)".

[0043] Sub-step three involves generating pixel tone information corresponding to the pixels of the initial low-light image block, based on the normalized pixel value information. In practice, the executing entity can generate pixel tone information based on the normalized pixel value information using color space conversion technology. As an example, the executing entity can input the updated red pixel value R, the updated green pixel value G, and the updated blue pixel value B into the following formula:

[0044]

[0045] The value of θ is in the range of [0, π]. When B is less than or equal to G, the pixel hue information can be θ, and when B is greater than G, the pixel hue information can be the difference between 2π and θ.

[0046] Sub-step four: Based on the above normalized pixel value information, generate pixel saturation information corresponding to the pixels of the initial low-light image block. In practice, the execution entity can generate pixel saturation information based on the above normalized pixel value information using color space conversion technology. As an example, the execution entity can input the updated red pixel value R, the updated green pixel value G, and the updated blue pixel value B into the following formula:

[0047]

[0048] When I equals 0, the pixel saturation information is 0; when I is not equal to 0, the pixel saturation information can be S, where S can be...

[0049] Sub-step five involves generating pixel brightness information corresponding to the pixels of the initial low-light image block, based on the aforementioned normalized pixel value information. In practice, the executing entity can generate pixel brightness information based on the aforementioned normalized pixel value information using color space conversion technology. As an example, the executing entity can determine the average of the updated red pixel value R, the updated green pixel value G, and the updated blue pixel value B as the pixel brightness information.

[0050] The second sub-step involves determining the hue information of each pixel in the generated initial low-light image block as the hue information of the initial low-light image block.

[0051] The third sub-step is to determine the saturation of each pixel in the generated initial low-light image block as the initial low-light image block saturation information.

[0052] The fourth sub-step involves determining the brightness information of each pixel in the generated initial low-light image block as the initial low-light image block brightness information.

[0053] The fifth sub-step involves inputting the initial low-light image patch's hue information, saturation information, and brightness information into a pre-trained image patch quality detection model to obtain image patch quality detection score information. This image patch quality detection model can be a neural network model (e.g., VGGNet (Visual Geometry Group Network) or Deep Residual Networks (ResNet)) that takes the initial low-light image patch's hue information, saturation information, and brightness information as input and outputs the image patch quality detection score information. The image patch quality detection score information can be a score of image quality. Optionally, the image patch quality detection model can be used to represent the correspondence between the initial low-light image patch's hue information, saturation information, brightness information, and image patch quality detection score information.

[0054] The fifth step is to generate image quality detection information based on the obtained quality detection scores of each image patch. In practice, the aforementioned executing entity can determine the average value of each score represented by the quality detection scores of each image patch as the image quality detection information.

[0055] The above-described technical solution, combined with its steps and related content, serves as an inventive point of this disclosure, addressing the technical problem of "waste of computing resources." Factors leading to this waste often include: When performing quality inspection on low-light images, the color features of RGB low-light images are highly complex. Changes in the values ​​of the red (R), green (G), and blue (B) channels simultaneously affect the color type, intensity, and brightness. Directly performing quality inspection on RGB low-light images using an image quality evaluation model increases the model's complexity, requiring more computational resources and thus wasting resources during the quality inspection process. Furthermore, the human visual system perceives color primarily based on hue, saturation, and brightness. However, the organization of the RGB color space does not directly match human visual perception. During image quality inspection, analysis results based on the RGB space may not align with subjective human perception. The human eye has an intuitive understanding of hue and saturation; for images where the model detects good quality but the human eye perceives poor quality, the image itself may have color combination or distribution issues, leading to lower accuracy in image quality inspection. Solving these factors can reduce the waste of computing resources. To achieve this effect, firstly, the area of ​​the initial low-light image is determined as the low-light image partitioning area. This yields the low-light image partitioning area used for dividing the low-light image. Then, in response to determining that the low-light image partitioning area is divisible by a preset partitioning area, the initial low-light image is divided into initial low-light image blocks with areas equal to the preset partitioning area and without overlapping images. This allows the initial low-light image to be divided into initial low-light image blocks. Next, in response to determining that the low-light image partitioning area is not divisible by the preset partitioning area, a sliding window partitioning process is performed on the initial low-light image based on a preset sliding window length, a preset sliding window width, a preset horizontal sliding step size, and a preset vertical sliding step size, resulting in initial low-light image blocks. This allows the initial low-light image to be divided into initial low-light image blocks even when the low-light image partitioning area is not divisible by the preset partitioning area. Next, for each of the obtained initial low-light image blocks, the following steps are performed: First, for each pixel in the initial low-light image block, the following steps are performed: First sub-step: The pixel value information of the pixel in the initial low-light image block is determined as the pixel value information to be converted. Thus, the pixel value information to be converted, used to generate normalized pixel value information, is obtained. Second sub-step: The information to be converted is normalized to obtain normalized pixel value information. Third sub-step: Based on the normalized pixel value information, pixel tone information corresponding to the pixel in the initial low-light image block is generated. Fourth sub-step: Based on the normalized pixel value information, pixel saturation information corresponding to the pixel in the initial low-light image block is generated.The fifth sub-step involves generating pixel brightness information corresponding to the pixels of the initial low-light image block based on the aforementioned normalized pixel value information. The second step involves determining the hue information of each pixel corresponding to the pixels of each initial low-light image block in the generated initial low-light image block as the initial low-light image block hue information. This yields the initial low-light image block hue information. The third step involves determining the saturation of each pixel corresponding to the pixels of each initial low-light image block in the generated initial low-light image block as the initial low-light image block saturation information. This yields the initial low-light image block saturation information characterizing the initial low-light image block saturation. The fourth step involves determining the brightness information of each pixel corresponding to the pixels of each initial low-light image block in the generated initial low-light image block as the initial low-light image block brightness information. This yields the initial low-light image block brightness information characterizing the initial low-light image block brightness. The fifth step involves inputting the initial low-light image patch's hue, saturation, and brightness information into a pre-trained image patch quality detection model to obtain image patch quality detection scoring information. This allows the separated hue (H), saturation (S), and brightness (I) information of the initial low-light image patch into the image patch quality detection model. This separation makes the data structure clearer, and the model doesn't need to consider the complex coupling relationships between RGB channels during processing, thus reducing data complexity and computational resource consumption during quality detection. Furthermore, the hue (H), saturation (S), and brightness (I) dimensions of the image patch are closely related to how the human eye perceives color. When judging image quality, the human eye primarily relies on the type of color (hue), vividness (saturation), and brightness (luminance). When HSI information is input into the quality evaluation model, the model's assessment of image quality more closely resembles human perception, improving the accuracy of image quality detection. Finally, based on the obtained quality detection scores of each image patch, image quality detection information is generated. This allows for the generation of image quality detection information for the corresponding low-light image. Furthermore, because the image data of the RGB low-light image patches is converted into initial low-light image patch hue (H), saturation (S), and luminance (I) information representing the image patch's hue (H), saturation (S), and luminance (I) during quality detection, the complex coupling relationships between RGB channels are not considered, reducing data complexity and computational resource consumption during quality detection.

[0056] Step 103: In response to determining that the image quality detection information meets the preset detection conditions, the initial low-light image is determined as a low-light image.

[0057] In some embodiments, the execution entity may determine the initial low-light image as a low-light image in response to determining that the image quality detection information meets a preset detection condition. The preset detection condition may be greater than or equal to a preset value (score).

[0058] Step 104: In response to determining that the image quality detection information does not meet the preset detection conditions, the exposure adjustment information corresponding to the image quality detection information is sent to the low-light camera device to adjust the sensitivity and exposure time of the low-light camera device, and the low-light image is acquired again through the adjusted low-light camera device.

[0059] In some embodiments, the execution entity may, in response to determining that the image quality detection information does not meet the preset detection conditions, send exposure adjustment information corresponding to the image quality detection information to the low-light camera device to adjust the sensitivity and exposure time of the low-light camera device, and then re-acquire a low-light image using the adjusted low-light camera device. The exposure adjustment information may be instruction information for controlling the sensitivity (ISO) and exposure time of the low-light camera device to change the image exposure level. For example, if the image quality detection information is "image quality is 3", then the exposure adjustment information corresponding to "image quality is 3" may be "{"device_id":"001", / / the number of the low-light camera device; "command":"exposure_adjustment", / / instruction type is exposure adjustment; "parameters":{"iso":800, / / increase the sensitivity to 800; "exposure_time":1 / 15 / / extend the exposure time to 1 / 15 second;}}".

[0060] Step 105: Obtain a normal light reference image.

[0061] In some embodiments, the execution entity can acquire a normal light reference image. In practice, the execution entity can acquire the normal light reference image via a wired connection or a wireless connection. The normal light reference image can be an image captured in a normal lighting environment.

[0062] It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other currently known or future wireless connection methods.

[0063] Step 106: Input the normal light reference image into the reference image feature extraction collaborative network of the pre-trained low-light image enhancement model to obtain the reference image feature information corresponding to the normal light reference image.

[0064] In some embodiments, the execution entity can input the normal light reference image into the reference image feature extraction cooperative network of a pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal light reference image. A schematic diagram of the model structure of the low-light image enhancement model is shown below. Figure 4 , Figure 5 As shown. The aforementioned low-light image enhancement model includes the aforementioned reference image feature extraction collaborative network and image enhancement association network. Figure 7 This is a schematic diagram of the network structure according to some embodiments of the reference image feature extraction collaborative network of this disclosure. The aforementioned reference image feature extraction collaborative network includes a reference image feature extraction module, pooling layers, and fully connected layers. The aforementioned reference image feature extraction module includes various convolutional modules, each of which includes a convolutional layer and an activation function. The aforementioned low-light image enhancement model can be a neural network model that takes a normal light reference image and a low-light image as input and enhances the low-light image as output. In some optional implementations of some embodiments, the executing entity can input the normal light reference image into the reference image feature extraction collaborative network of the pre-trained low-light image enhancement model through the following steps to obtain reference image feature information corresponding to the normal light reference image:

[0065] The first step involves inputting the aforementioned normal light reference image into the reference image feature extraction module of the aforementioned reference image feature extraction collaborative network to obtain initial feature information of the reference image. The reference image feature extraction module includes various convolutional modules, and the reference image feature extraction collaborative network includes the aforementioned reference image feature extraction module, pooling layers, and fully connected layers. The aforementioned reference image feature extraction collaborative network can be a neural network for feature extraction from the normal light reference image. Each convolutional module includes a convolutional layer and an activation function. The output information of the convolutional layer points can be input to the activation function layer. The initial feature information of the reference image can be the feature vector or feature map obtained by feature extraction from the normal light reference image.

[0066] The initial feature information of the reference image is input into the pooling layer to obtain the pooled feature information of the reference image. This pooled feature information can be a feature vector or feature map obtained by pooling the initial feature information of the reference image.

[0067] The pooling feature information of the aforementioned reference image is input into the fully connected layer to obtain reference image feature information corresponding to the aforementioned normal light reference image. This reference image feature information can be a higher-level abstract feature vector or a higher-level feature map obtained by processing the pooling feature information of the reference image through the fully connected layer.

[0068] Step 107: Input the feature information of the low-light image and the reference image into the image enhancement association network to obtain the enhanced low-light image of the corresponding low-light image.

[0069] In some embodiments, the execution entity may input the feature information of the low-light image and the reference image into the image enhancement association network to obtain an enhanced low-light image corresponding to the low-light image.

[0070] In some optional implementations of certain embodiments, the execution entity may input the feature information of the low-light image and the reference image into the image enhancement association network through the following steps to obtain an enhanced low-light image corresponding to the low-light image:

[0071] The first step involves inputting the aforementioned low-light images into the feature extraction module sequence of the image enhancement association network to obtain the extracted feature information for each low-light image. A schematic diagram of the network structure of the image enhancement association network is shown below. Figure 6 As shown. The image enhancement association network described above includes the aforementioned feature extraction module sequence and feature fusion module sequence. Each feature extraction module in the feature extraction module sequence includes a convolutional layer and a self-attention block. Each feature fusion module in the feature fusion module sequence corresponds to at least one feature extraction module in the feature extraction module sequence. The self-attention block can be a Transformer block. The first feature extraction module in the feature extraction module sequence is configured to extract features from the low-light image, extracting feature vectors that represent the contextual information of the image. Each feature fusion module in the feature fusion module sequence includes a convolutional layer and a Transformer block. The feature fusion module can be a decoder. The feature extraction module can be an encoder.

[0072] The second step involves inputting at least one low-light image extraction feature from at least one feature extraction module corresponding to the aforementioned feature fusion module, the output information of the previous feature fusion module in the sequence, and the reference image feature information into the aforementioned feature fusion module to obtain fused feature information. This fused feature information is then input into the next feature fusion module in the sequence. The fused feature information can be a comprehensive representation of feature information from multiple sources (here, at least one low-light image extraction feature, the output information of the previous feature fusion module, and the reference image feature information). This fused feature information can be a feature map combining features from the low-light image, accumulated information during the feature fusion process, and feature information from the normal light image.

[0073] The third step involves determining the image corresponding to the fused feature information output by the last feature fusion module in the aforementioned feature fusion module sequence as the enhanced low-light image corresponding to the aforementioned low-light image. The fused feature information output by the last feature fusion module in the aforementioned feature fusion module sequence can characterize the enhanced low-light image after mapping the multi-level feature information of the normal light reference image onto the low-light image. In practice, the executing entity can input the fused feature information output by the last feature fusion module into a preset decoder or preset generator network to map the fused feature information back into the image space, obtaining the image corresponding to the fused feature information.

[0074] In some optional implementations of certain embodiments, the execution entity may input the low-light image into the feature extraction module sequence of the image enhancement association network through the following steps to obtain the extracted feature information of each low-light image:

[0075] First, based on the feature extraction module sequence described above, perform the following steps:

[0076] Sub-step one: Identify each feature extraction module in the above feature extraction module sequence, except for the first feature extraction module, as the target feature extraction module.

[0077] Sub-step two: Determine the target feature extraction module located in the sequence following the first feature extraction module.

[0078] In sub-step three, for the first feature extraction module in the above feature extraction module sequence, the low-light image is input into the first feature extraction module to obtain initial low-light image extraction feature information corresponding to the low-light image. In practice, the convolutional layer included in the feature extraction module can perform feature extraction on the low-light image to obtain a feature map representing the low-light image. Then, the self-attention block included in the feature extraction module can receive the feature map representing the low-light image. For each position (pixel or group of pixels) in the low-light image feature map, its feature value is adjusted according to the calculated attention weight (for example, if the attention weight corresponding to a certain pixel position is high, then its feature value in the output feature map will be relatively enhanced; conversely, if the weight is low, the feature value will be relatively suppressed). The resulting feature map, reweighted by the attention weight, serves as the initial low-light image extraction feature information.

[0079] Sub-step four involves inputting the feature information extracted from the initial low-light image into at least one feature fusion module corresponding to the first feature extraction module.

[0080] Sub-step five involves inputting the initial low-light image extracted feature information into the target feature extraction module. Each target feature extraction module in the feature extraction module sequence is configured to receive the output information of the previous feature extraction module in the sequence, and to output the initial low-light image extracted feature information. The previous feature extraction module of the first target feature extraction module in the feature extraction module sequence can be the first feature extraction module in the sequence. For each target feature extraction module in the feature extraction module sequence other than the first target feature extraction module, the previous feature extraction module can be the previous target feature extraction module in the sequence.

[0081] Sub-step six: Determine the initial low-light image extraction feature information output by the first feature extraction module and each target feature extraction module in the feature extraction module sequence as the low-light image extraction feature information.

[0082] Optionally, the above low-light image enhancement model can be trained through the following steps:

[0083] The first step is to obtain a sample set. This sample set includes low-light reference image data and corresponding normal-light images of the target sample. Specifically, the low-light reference image data includes both low-light and normal-light reference images.

[0084] The second step involves performing the following training steps based on the sample set:

[0085] The first sub-step involves inputting the low-light reference image data of at least one sample in the sample set into the initial neural network to obtain the sample prediction normal light image corresponding to each of the at least one sample.

[0086] The second sub-step involves comparing the predicted normal light image of each sample in the at least one set of samples with the corresponding target normal light image. In practice, the execution entity can use the cross-entropy loss function to determine the difference between the predicted normal light image of each sample in the at least one set of samples and the corresponding target normal light image.

[0087] The third sub-step involves determining whether the initial neural network has reached the preset optimization objective based on the comparison results. This optimization objective can be that the cross-entropy loss function value is less than or equal to a preset value.

[0088] The fourth sub-step is to use the initial neural network as the trained low-light image enhancement model in response to the determination that the initial neural network has achieved the above optimization objective.

[0089] The fifth sub-step involves adjusting the network parameters of the initial neural network in response to the determination that the initial neural network has not achieved the above optimization objective. A sample set is then created using unused samples, and the adjusted initial neural network is used as the new initial neural network. The training steps described above are then executed again. As an example, the back propagation algorithm (BP algorithm) and gradient descent methods (such as mini-batch gradient descent) can be used to adjust the network parameters of the initial neural network.

[0090] The various embodiments of this disclosure have the following beneficial effects: the low-light image enhancement methods of some embodiments of this disclosure improve the image quality and image clarity of the enhanced low-light image. Specifically, the reason for the poor image quality and image clarity of the enhanced low-light image is that: directly enhancing the low-light image captured by the low-light camera device does not optimize the acquisition process based on the low-light image captured in the low-light environment. The original image captured without optimization may contain a lot of noise, which is also amplified during the enhancement process, making the noise in the enhanced low-light image more obvious, resulting in a more grainy and coarser image quality. At the same time, while directly enhancing the low-light image captured by the low-light camera device can improve the brightness and contrast of the low-light image to some extent, the quality of the enhanced image may still be limited due to the poor quality of the original image, resulting in a relatively blurry enhanced low-light image. Based on this, the low-light image enhancement method of some embodiments of this disclosure first acquires an initial low-light image using a low-light camera device. Thus, an initial low-light image can be obtained to determine whether to re-acquire the low-light image. Then, image quality detection processing is performed on the initial low-light image to obtain image quality detection information. This allows for image quality detection of the initial low-light image to generate image quality detection information. This image quality detection information characterizes the image quality acquired by the low-light camera device in a low-light environment without optimized acquisition process. Subsequently, in response to determining that the image quality detection information meets preset detection conditions, the initial low-light image is identified as a low-light image. Thus, given that the image quality detection information of the initial low-light image meets the preset detection conditions, an initial low-light image with relatively good image quality can be identified as a low-light image. Then, in response to determining that the image quality detection information does not meet the preset detection conditions, exposure adjustment information corresponding to the image quality detection information is sent to the low-light camera device to adjust the sensitivity and exposure time of the low-light camera device, and the low-light image is acquired again using the adjusted low-light camera device. Thus, even when the image quality of the initial low-light image is poor (i.e., the image quality detection information does not meet the preset detection conditions), the sensitivity and exposure time of the low-light camera device can be adjusted to optimize the parameters of the low-light camera device and improve the acquisition process. After quality inspection, adjusting the ISO and exposure time allows the low-light camera to better adapt to low-light environments, capturing relatively high-quality low-light images and reducing noise introduction. Then, a normal-light reference image is acquired. This provides a normal-light reference image for generating reference image feature information. Next, this normal-light reference image is input into the reference image feature extraction collaborative network of a pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal-light reference image. The low-light image enhancement model includes the reference image feature extraction collaborative network and an image enhancement association network.Therefore, reference image feature information of a normal light reference image can be extracted through a reference image feature extraction collaborative network. Then, the aforementioned low-light image and the aforementioned reference image feature information are input into the aforementioned image enhancement association network to obtain an enhanced low-light image corresponding to the aforementioned low-light image. Thus, the low-light image can be enhanced based on the reference image feature information through the image enhancement association network to obtain an enhanced low-light image. Furthermore, because the initial low-light image acquired by the low-light camera device was quality-checked before enhancement, and the sensitivity and exposure time of the low-light camera device were adjusted to optimize the acquisition process, the parameters of the low-light camera device were optimized. Adjusting the sensitivity and exposure time after quality check allows the low-light camera device to better adapt to low-light environments, acquire relatively high-quality low-light images, reduce the introduction of noise in the low-light image itself, and improve the image quality and sharpness of the enhanced low-light image.

[0091] Further reference Figure 2 As an implementation of the methods shown in the figures, this disclosure provides some embodiments of a low-light image enhancement apparatus, which are similar to... Figure 1 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0092] like Figure 2As shown, a low-light image enhancement device 200 in some embodiments includes: a first acquisition unit 201, an image quality detection unit 202, a determination unit 203, a second acquisition unit 204, an acquisition unit 205, a first input unit 206, and a second input unit 207. The first acquisition unit 201 is configured to acquire an initial low-light image using a low-light camera; the image quality detection unit 202 is configured to perform image quality detection processing on the initial low-light image to obtain image quality detection information; the determination unit 203 is configured to determine the initial low-light image as a low-light image in response to determining that the image quality detection information meets preset detection conditions; the second acquisition unit 204 is configured to send exposure adjustment information corresponding to the image quality detection information to the low-light camera in response to determining that the image quality detection information does not meet preset detection conditions, so as to adjust the sensitivity and exposure time of the low-light camera. The low-light image is captured again using the adjusted low-light camera device; the acquisition unit 205 is configured to acquire a normal light reference image; the first input unit 206 is configured to input the normal light reference image into the reference image feature extraction collaborative network of the pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal light reference image, wherein the low-light image enhancement model includes the reference image feature extraction collaborative network and the image enhancement association network; the second input unit 207 is configured to input the low-light image and the reference image feature information into the image enhancement association network to obtain an enhanced low-light image corresponding to the low-light image.

[0093] It is understandable that the units described in the device 200 are related to the reference. Figure 1 The steps in the method described above correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the device 200 and the units contained therein, and will not be repeated here.

[0094] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0095] like Figure 3As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0096] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0097] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0098] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0099] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0100] Computer-readable media may be contained within an electronic device or may exist independently of the electronic device. A computer-readable medium carries one or more programs that, when executed by an electronic device, cause the electronic device to: acquire an initial low-light image using a low-light camera; perform image quality detection processing on the initial low-light image to obtain image quality detection information; in response to determining that the image quality detection information meets preset detection conditions, identify the initial low-light image as a low-light image; in response to determining that the image quality detection information does not meet preset detection conditions, send exposure adjustment information corresponding to the image quality detection information to the low-light camera to adjust the sensitivity and exposure time of the low-light camera, and acquire the low-light image again using the adjusted low-light camera; acquire a normal light reference image; input the normal light reference image into a reference image feature extraction collaborative network of a pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal light reference image, wherein the low-light image enhancement model includes the reference image feature extraction collaborative network and an image enhancement association network; input the low-light image and the reference image feature information into the image enhancement association network to obtain an enhanced low-light image corresponding to the low-light image.

[0101] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0102] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0103] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including a first acquisition unit, an image quality detection unit, a determination unit, a second acquisition unit, an acquisition unit, a first input unit, and a second input unit. The names of these units do not necessarily limit the specific unit; for example, the first acquisition unit may also be described as "a unit for acquiring an initial low-light image using a low-light camera device."

[0104] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0105] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of technical features, but should also cover other technical solutions formed by arbitrary combinations of technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A low-light image enhancement method, comprising: Initial low-light images are captured using a low-light camera device; The initial low-light image is subjected to image quality detection processing to obtain image quality detection information; In response to determining that the image quality detection information meets the preset detection conditions, the initial low-light image is determined as a low-light image; In response to determining that the image quality detection information does not meet the preset detection conditions, exposure adjustment information corresponding to the image quality detection information is sent to the low-light camera device to adjust the sensitivity and exposure time of the low-light camera device, and to re-acquire a low-light image through the adjusted low-light camera device; Acquire a normal light reference image; The normal light reference image is input into the reference image feature extraction collaborative network of the pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal light reference image. The low-light image enhancement model includes the reference image feature extraction collaborative network and the image enhancement association network. The low-light image and the reference image feature information are input into the image enhancement association network to obtain an enhanced low-light image corresponding to the low-light image. The step of inputting the low-light image and the reference image feature information into the image enhancement association network to obtain the enhanced low-light image corresponding to the low-light image includes: The low-light image is input into the feature extraction module sequence of the image enhancement association network to obtain feature information extracted from each low-light image. The image enhancement association network includes the feature extraction module sequence and the feature fusion module sequence. Each feature extraction module in the feature extraction module sequence includes a convolutional layer and a self-attention block. Each feature fusion module in the feature fusion module sequence corresponds to at least one feature extraction module in the feature extraction module sequence. For each feature fusion module in the feature fusion module sequence, at least one low-light image extraction feature information output by at least one feature extraction module corresponding to the feature fusion module, the output information of the previous feature fusion module in the feature fusion module sequence, and the reference image feature information are input to the feature fusion module to obtain fused feature information, and the fused feature information is input to the next feature fusion module in the feature fusion module sequence. The image corresponding to the fused feature information output by the last feature fusion module in the feature fusion module sequence is determined as the enhanced low-light image corresponding to the low-light image.

2. The method according to claim 1, wherein, The step of inputting the normal light reference image into the reference image feature extraction collaborative network of the pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal light reference image includes: The normal light reference image is input into the reference image feature extraction module of the reference image feature extraction collaborative network to obtain the initial feature information of the reference image. The reference image feature extraction module includes various convolutional modules, and the reference image feature extraction collaborative network includes the reference image feature extraction module, pooling layer, and fully connected layer. The initial feature information of the reference image is input into the pooling layer to obtain the pooling feature information of the reference image; The pooling feature information of the reference image is input into the fully connected layer to obtain the reference image feature information corresponding to the normal light reference image.

3. The method according to claim 1, wherein, The step of inputting the low-light image into the feature extraction module sequence of the image enhancement association network to obtain feature information of each low-light image includes: Based on the feature extraction module sequence, perform the following steps: Each feature extraction module in the feature extraction module sequence, except for the first feature extraction module, is identified as a target feature extraction module. Determine the target feature extraction module located at the next position in the feature extraction module sequence after the first feature extraction module; For the first feature extraction module in the feature extraction module sequence, the low-light image is input into the first feature extraction module to obtain the initial low-light image extraction feature information corresponding to the low-light image; The feature information extracted from the initial low-light image is input into at least one feature fusion module corresponding to the first feature extraction module; The initial low-light image extracted feature information is input to the target feature extraction module, wherein each target feature extraction module in the feature extraction module sequence is configured to receive the output information of the previous feature extraction module in the feature extraction module sequence, so as to output the initial low-light image extracted feature information; The feature information extracted from each initial low-light image output by the first feature extraction module and each target feature extraction module in the feature extraction module sequence is determined as the feature information extracted from each low-light image.

4. The method according to claim 1, wherein, The low-light image enhancement model is trained through the following steps: Obtain a sample set, wherein the samples in the sample set include sample low-light reference image data and sample target normal light images corresponding to the sample low-light reference image data, wherein the sample low-light reference image data includes sample low-light images and sample normal light reference images; Perform the following training steps based on the sample set: The low-light reference image data of at least one sample in the sample set is input into the initial neural network to obtain the sample prediction normal light image corresponding to each of the at least one sample; Compare the predicted normal light image of each sample in the at least one sample with the corresponding target normal light image of the sample. Based on the comparison results, determine whether the initial neural network has achieved the preset optimization objective; In response to determining that the initial neural network has achieved the optimization objective, the initial neural network is used as the trained low-light image enhancement model; In response to the determination that the initial neural network has not achieved the optimization objective, the network parameters of the initial neural network are adjusted, and a sample set is formed using unused samples. The adjusted initial neural network is then used as the initial neural network, and the training steps are performed again.

5. A low-light image enhancement device, comprising: The first acquisition unit is configured to acquire an initial low-light image using a low-light camera device; An image quality detection unit is configured to perform image quality detection processing on the initial low-light image to obtain image quality detection information; The determining unit is configured to determine the initial low-light image as a low-light image in response to determining that the image quality detection information meets a preset detection condition; The second acquisition unit is configured to, in response to determining that the image quality detection information does not meet the preset detection conditions, send exposure adjustment information corresponding to the image quality detection information to the low-light camera device to adjust the sensitivity and exposure time of the low-light camera device, and to acquire a low-light image again through the adjusted low-light camera device; The acquisition unit is configured to acquire a normal light reference image; The first input unit is configured to input the normal light reference image into the reference image feature extraction collaborative network of a pre-trained low-light image enhancement model to obtain reference image feature information corresponding to the normal light reference image, wherein the low-light image enhancement model includes the reference image feature extraction collaborative network and the image enhancement association network. The second input unit is configured to input the feature information of the low-light image and the reference image into the image enhancement association network to obtain an enhanced low-light image corresponding to the low-light image. The step of inputting the feature information of the low-light image and the reference image into the image enhancement association network to obtain the enhanced low-light image corresponding to the low-light image includes: inputting the low-light image into a feature extraction module sequence of the image enhancement association network to obtain extracted feature information for each low-light image. The image enhancement association network includes the feature extraction module sequence and a feature fusion module sequence. Each feature extraction module in the feature extraction module sequence includes a convolutional layer and a self-attention block. Each feature fusion module in the feature fusion module sequence corresponds to at least one feature extraction module in the feature extraction module sequence. Each feature extraction module in the feature extraction module sequence corresponds to at least one feature fusion module in the feature fusion module sequence. For each feature fusion module in the feature fusion module sequence, at least one low-light image extraction feature information output by at least one feature extraction module corresponding to the feature fusion module, the output information of the previous feature fusion module in the feature fusion module sequence, and the reference image feature information are input to the feature fusion module to obtain fused feature information. The fused feature information is then input to the next feature fusion module in the feature fusion module sequence. The image corresponding to the fused feature information output by the last feature fusion module in the feature fusion module sequence is determined as the enhanced low-light image corresponding to the low-light image.

6. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 4.

7. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 4.

Images