Image Enhancement Method and System Based on Gamma Correction Prior-Driven Attention Mechanism

Abstract

The present invention discloses an image enhancement method and system based on a gamma-corrected prior-driven attention mechanism, relating to the field of optics. The method comprises the following steps: constructing a low-light image enhancement neural network based on a pre-configured network structure; training and evaluating the low-light image enhancement neural network based on a pre-selected loss function, training dataset, and evaluation metrics; and performing ablation experiments on the trained low-light image enhancement neural network, sequentially adjusting the number of extended gamma-corrected attention modules, channel prior modules, and loss functions to obtain an enhanced image. The present invention proposes a new model as an attention mechanism based on a theoretical formula, and proposes stacked and modular attention modules to focus on image details. This enhancement achieves lightweight yet advanced performance, maintaining strong efficiency and stable operation.

Description

Technical Field

[0001] The present invention relates to the field of optics, and in particular to an image enhancement method and system based on a gamma correction prior-driven attention mechanism. Background Art

[0002] Low-light image enhancement is a key challenge in computer vision. Currently, two main approaches are commonly used to address low-light image enhancement: histogram equalization and Retinex theory. Histogram equalization enhances contrast by redistributing grayscale values. Retinex theory, on the other hand, separates an image into reflectance and illumination components to improve reflectance and overall image quality.

[0003] With technological advancements, various deep learning-based methods have been proposed in recent years to improve the quality of low-light images. However, these methods often require extensive computational resources, which limits their practical application on real-world devices. Therefore, designing lightweight and efficient image enhancement techniques is crucial.

[0004] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0005] In response to the problems in the related art, the present invention proposes an image enhancement method and system based on a gamma correction prior-driven attention mechanism to overcome the above-mentioned technical problems existing in the existing related art.

[0006] To this end, the specific technical solutions adopted in the present invention are as follows:

[0007] According to one aspect of the present invention, an image enhancement method based on a gamma correction prior-driven attention mechanism is provided, the image enhancement method comprising the following steps:

[0008] S1. Build a low-light image enhancement neural network based on a pre-configured network structure.

[0009] S2. Train and evaluate the low-light image enhancement neural network based on the pre-selected loss function, training dataset, and evaluation metrics;

[0010] S3. Based on the trained low-light image enhancement neural network, ablation experiments are conducted on the extended gamma correction attention module, the number of channel prior modules, and the loss function to obtain the enhanced image;

[0011] The low-light image enhancement neural network includes a channel prior module for fusing original RGB channel features and an extended gamma correction attention module for estimating global features.

[0012] Optionally, constructing a low-light image enhancement neural network based on a preconfigured network structure, wherein the low-light image enhancement neural network includes a channel prior module and an extended gamma correction attention module, comprises the following steps:

[0013] S11, channel prior module obtained based on Retinex theory and pre-configured atmospheric scattering model;

[0014] S12. Gamma correction is introduced into the attention process, and combined with the channel prior module, the optimal gamma correction of the feature channel is estimated using global features to obtain the extended gamma correction attention module.

[0015] Optionally, the obtaining of a channel priori module based on Retinex theory and a pre-configured atmospheric scattering model comprises the following steps:

[0016] S111, couple the Retinex theory with the pre-configured atmospheric scattering model and transform it into a parameter learning model;

[0017] S112. Use the convolutional network layer and the residual network layer to reconstruct the network structure expression of the parameter learning model to obtain the channel prior module.

[0018] Optionally, the expressions of the channel prior module and the extended gamma correction attention module are:

[0019] ;

[0020] ;

[0021] Where, represents the mapped input tensor with matching channels, represents the complement of atmospheric light intensity, represents atmospheric transmission, represents the input feature map, represents the channel prior characteristics, Represents gamma-corrected features and original features The attention features established, Indicates that based on the gamma correction feature, Represents the original feature of attention.

[0022] Optionally, the loss function includes L1 loss, perceptual loss, HDR loss and SSIM loss;

[0023] When the low-light image enhancement neural network is trained using matching data, the training data set includes a League of Legends v1 (LOLv1) data set and a League of Legends v2 (LOLv2) data set, the LOLv2 data set includes a real capture subset and a synthetic subset, and the evaluation metrics include peak signal-to-noise ratio, structural similarity, and learnable perceptual image block similarity;

[0024] When the low-light image enhancement neural network is trained using unmatched data, the training data set includes a low-light image data set enhanced by illumination map estimation, a data set of a power-limited contrast enhancement technology for emissive displays based on histogram equalization, a naturalness enhancement algorithm data set for non-uniform illumination images, a cracked image enhancement and tone mapping algorithm data set, and a contrast enhancement technology data set based on hierarchical difference representation, and the evaluation index includes a natural image quality evaluator.

[0025] Optionally, the step of sequentially performing ablation experiments on the number of extended gamma correction attention modules, channel prior modules, and loss functions based on the trained low-light image enhancement neural network includes the following steps:

[0026] S31. Based on gamma correction and attention mechanism, an ablation experiment is conducted on the extended gamma correction attention module.

[0027] S32, based on the low-light image enhancement neural network, an ablation experiment was conducted by changing the number of channel prior modules;

[0028] S33. Based on the default settings of the low-light image enhancement neural network, an ablation experiment is performed on the loss function.

[0029] Optionally, the ablation experiment on the extended gamma correction attention module based on the gamma correction combined with the attention mechanism includes the following steps:

[0030] S311. Establish an attention mechanism based on gamma correction;

[0031] S312. Integrate gamma correction from the global branch to the local branch and perform ablation experiments on the extended gamma correction attention module.

[0032] Optionally, performing an ablation experiment based on the low-light image enhancement neural network by changing the number of channel prior modules includes the following steps:

[0033] S321, based on the low-light image enhancement neural network, gradually changing the number of channel prior modules;

[0034] S322. Evaluate the model performance under each configuration separately and record the number of parameters and computational cost.

[0035] Optionally, performing an ablation experiment on the loss function based on the default setting of the low-light image enhancement neural network includes the following steps:

[0036] S331. Based on the default settings of the low-light image enhancement neural network, ablation experiments are performed on the loss function respectively;

[0037] S332. Combined with the loss function after the ablation experiment, the best overall monitoring performance is obtained.

[0038] According to another aspect of the present invention, an image enhancement system based on a gamma correction prior-driven attention mechanism is also provided, the image enhancement system comprising a neural network construction module, a neural network training module and an ablation experiment module;

[0039] A neural network building module for building a low-light image enhancement neural network based on a pre-configured network structure;

[0040] A neural network training module is used to train and evaluate the low-light image enhancement neural network based on a pre-selected loss function, training dataset, and evaluation metrics;

[0041] The ablation experiment module is used to perform ablation experiments on the extended gamma correction attention module, the number of channel prior modules, and the loss function based on the trained low-light image enhancement neural network to obtain the enhanced image.

[0042] The beneficial effects of the present invention are:

[0043] This paper proposes a new model CPGA-Net+, which uses it as an attention mechanism based on a theoretical formula and proposes a stacked and modular attention module to focus on the details of the image. In addition, gamma correction is integrated into the local branch to create a plug-in attention module for each CP block. This enhancement enables lightweight but advanced performance, maintaining strong efficiency and stable operation on devices with limited computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 is a flowchart of an image enhancement method based on a gamma correction prior-driven attention mechanism according to an embodiment of the present invention;

[0046] Figure 22 is a block diagram of an image enhancement system based on a gamma correction prior-driven attention mechanism according to an embodiment of the present invention;

[0047] Figure 3 1. CPGA-Net+ principle diagram in an image enhancement method based on a gamma correction prior-driven attention mechanism according to an embodiment of the present invention;

[0048] Figure 4 1 is a design block diagram of a channel prior block (CP block) in an image enhancement method based on a gamma correction prior driven attention mechanism according to an embodiment of the present invention;

[0049] Figure 5 2 is a schematic structural diagram of an intersection-aware adaptive fusion module (IAAF) in an image enhancement method based on a gamma-corrected prior-driven attention mechanism according to an embodiment of the present invention.

[0050] In the picture:

[0051] 1. Neural network construction module; 2. Neural network training module; 3. Ablation experiment module. DETAILED DESCRIPTION

[0052] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0053] According to an embodiment of the present invention, an image enhancement method and system based on a gamma correction prior-driven attention mechanism are provided.

[0054] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 、 Figure 3-Figure 5 As shown, according to one embodiment of the present invention, an image enhancement method based on a gamma correction prior-driven attention mechanism is provided, and the image enhancement method includes the following steps:

[0055] S1. Based on a preconfigured network structure, a low-light image enhancement neural network is constructed; wherein the low-light image enhancement neural network includes a channel prior module for fusing original RGB channel features and an extended gamma correction attention module for estimating global features.

[0056] In one embodiment, constructing a low-light image enhancement neural network based on a preconfigured network structure, wherein the low-light image enhancement neural network includes a channel prior module and an extended gamma correction attention module, comprises the following steps:

[0057] S11. Obtain channel prior module based on Retinex theory and pre-configured atmospheric scattering model.

[0058] In one embodiment, the obtaining of a channel priori module based on Retinex theory and a preconfigured atmospheric scattering model comprises the following steps:

[0059] S111, couple the Retinex theory with the pre-configured atmospheric scattering model and transform it into a parameter learning model;

[0060] S112. Use the convolutional network layer and the residual network layer to reconstruct the network structure expression of the parameter learning model to obtain the channel prior module.

[0061] S12. Gamma correction is introduced into the attention process, and combined with the channel prior module, the optimal gamma correction of the feature channel is estimated using global features to obtain the extended gamma correction attention module.

[0062] In one embodiment, the expressions of the channel prior module and the extended gamma correction attention module are:

[0063] ;

[0064] ;

[0065] Where, represents the mapped input tensor with matching channels, represents the complement of atmospheric light intensity, represents atmospheric transmission, represents the input feature map, represents the channel prior characteristics, Represents gamma-corrected features and original features The attention features established, Indicates that based on the gamma correction feature, Represents the original feature of attention.

[0066] It is important to explain that, leveraging a theory-based structure and gamma correction priors, we extend CPGA-Net by reformulating the theoretical equations as an attention mechanism and transitioning the gamma estimation module from global to local processing. We refer to this network as CPGA-Net+, and this model achieves state-of-the-art image quality and efficiency, making it a lightweight and practical solution for real-world applications.

[0067] The main features of this model are as follows:

[0068] Atmospheric scattering driven attention module: This paper introduces the atmospheric scattering model as attention into the local network, called the "channel prior block". It improves the capture of channel prior information to better preserve and refine image structure and details.

[0069] Plug-in attention module with gamma correction: Gamma correction is integrated into the local branch, converting it into an attention module similar to CPGA-Net. This maximizes the efficiency of environmental factors and improves image quality through the plug-in attention mechanism.

[0070] A gamma correction prior driven attention mechanism low-light image enhancement neural network is designed, which mainly includes a channel prior module CP module and an extended gamma correction attention module CPGA module.

[0071] Among them, regarding the channel prior module CP module, an attention mechanism called channel-prior block (CP block) is proposed, which modularizes the relationship between the two equations of Retinex theory and atmospheric scattering model into a system form, and reorganizes the module with convolution and ResBlock, while fusing the features of the original RGB channels at each step to simplify the module into an attention block design, which helps to activate the feature map to align with the original channel.

[0072] The CP module, short for Channel-Prior Block, is a module inspired by channel priors and driven by an atmospheric scattering model. This module couples the Retinex theory and the atmospheric scattering model, transforming them into a parameter-learning network. The convolutional network layer CNNBlock and the residual network layer ResBlock are used to reconstruct the network structure that integrates the Retinex and atmospheric scattering models. Furthermore, the original RGB channel features are integrated at each step to simplify the module into an attention block design.

[0073] There is a strong relationship between Retinex theory and the atmospheric scattering model (ATSM), and these two equations can be expressed in terms of each other by reformulating them. Retinex theory assumes that the received image can be decomposed into illumination and reflectance components as follows:

[0074] ;

[0075] Where, express The reflectance component of the pixel position, express The perceived image at pixel location , express The lighting component of the pixel position.

[0076] On the other hand, the atmospheric scattering model is defined as follows:

[0077] ;

[0078] Where, I ( x ) represents the input image, J ( x ) represents a fog-free image, t ( x ) represents atmospheric transmission, A ( x ) represents the atmospheric light intensity.

[0079] Due to low-light images I ( x ) can be regarded as 1− L ( x ), J ( x ) can be regarded as 1− R ( x ). Therefore, it can be rewritten as follows:

[0080] ;

[0081] Where, R ( x ) represents the reflectance component of the image, L ( x ) represents the illumination component of the image, The complement of the atmospheric light intensity used to adjust the haze or scattering effect.

[0082] in =1- A ( x ), it is worth noting that if =0, the equation is consistent with the Retinex theory.

[0083] This paper extends this idea and proposes an attention mechanism called Channel-Prior Block (CP Block), which modularizes the relationship between the above two equations into a system form and reorganizes the module with convolution and ResBlock (residual block), while fusing the features of the original RGB channels at each step to simplify the module into an attention block design, which helps to activate the feature map to align with the original channel.

[0084] First, three channel priors are selected as the input for t estimation: bright channel prior (BCP), dark channel prior (DCP), and luminance channel (Y component from YCbCr color space). They can be defined as:

[0085] ;

[0086] ;

[0087] ;

[0088] Where, represents the bright channel prior, represents the color channels of the input image, represents the dark channel prior, represents the brightness channel, represents the red channel in the image, represents the green channel in the image, Represents the blue channel in the image.

[0089] in, C Represents the input image I Color channels C ,These common features represent the brightness changes under different environments, such as Figure 4 As shown, Figure 4 Here, Channel-Priors stands for channel priors, Original Channel stands for original channel, concat stands for concatenation, GELU is the activation function, and ResBlock stands for residual block. These common features represent brightness variations under different environments and have been widely used in traditional methods. The combination of channel priors shows sensitivity to brightness, which is an important cue for characterizing atmospheric transmittance and guiding enhancement.

[0090] Furthermore, in order to convert it into a high-dimensional processing module instead of keeping the original channels, the present invention simplifies the brightness channel into a more basic representation - specifically, the average value of the channel, similar to the concept of intensity (I component in HSI color space), which also represents image brightness. The channel prior is described as:

[0091] ;

[0092] Where, Indicates the maximum value of the input channel, Indicates the minimum value of the input channel, represents the average value of the input channel, concat Indicates that the values extracted from the max, min, and mean of the input channel are sequentially connected into a new feature representation. F represents the input feature map, represents the channel prior characteristics.

[0093] Among them, if C consists of RGB channels, then FEqual to I, so that the channel prior is simplified to the max, min and mean of the input channel, making the attention module more sensitive to the overall brightness control of the image.

[0094] For capturing detailed features and reconstructing images Based on the estimation of the proposed method, the present invention redesigns it into a micro-U-Net-based architecture, in which the encoder and decoder paths are connected via skip connections. This design effectively captures and preserves multi-scale spatial information, enhancing the model's ability to reconstruct fine details in the image. Furthermore, considering lightweight efficiency, the present invention samples the input only once, reducing computational complexity while maintaining sufficient feature extraction capabilities. This approach ensures that the model remains effective and suitable for real-time applications or scenarios with limited computing resources, without significantly compromising the quality of the reconstructed image.

[0095] After obtaining t and After estimating , the features can be reconstructed using the atmospheric scattering model combined with the low-light image as an attention module that is sensitive to brightness changes in the scene, which leads to the proposed atmospheric scattering driven attention, which is formulated as follows:

[0096] ;

[0097] Where, L ' represents the mapped input tensor with matching channels, Represents attention features.

[0098] In addition, we design a CPGA module, which is essentially an extension of the gamma correction attention module. This module directly incorporates gamma correction into the attention process, adapting this operation to the function within the CP module, enabling it to be used as a global feature to estimate the optimal gamma correction for each feature channel.

[0099] The CPGA-Net model not only takes advantage of the characteristics of low-light images, but also combines gamma correction based on IAT. Gamma correction is a simple technique that adjusts all pixels through point-by-point exponential operation, as shown below:

[0100] ;

[0101] Where, γ Indicates the gamma value that controls the degree of correction, r represents the enhanced input image, s Represents the output image.

[0102] in γ is the gamma value that controls the degree of correction and enhances the input image r To generate the output image sThese methods combine independent regression branches with enhanced models to better estimate the gamma value. In addition, the complexity of gamma value estimation will complicate the training objective and make the process prone to divergence. Considering these factors, such as Figure 5 As shown in the figure, ResBlock represents the residual block. An intersection-aware adaptive fusion module (IAAF) is proposed, which is expressed as follows:

[0103] ;

[0104] Where, represents the enhanced image after being processed by the cross-sensing adaptive fusion module, IAAF represents the intersection perception adaptive fusion module, Indicates that R and All independent information in Representation extraction and The common information between represents a gamma-corrected image, R Represents the original image.

[0105] Among them, the enhanced image is by R and created by combining them together while removing any overlapping elements, Represents an estimate of finding the intersection of similar features between the gamma-corrected image and the original image.

[0106] First, downsampling the image resolution at a limited scale does not significantly affect the performance while greatly reducing the computational cost. Therefore, the gamma value estimation process is redesigned, such as Figure 3 As shown, Figure 3 Pre-processing stands for pre-processing module, CP stands for channel-prior module, Fusion Strategy stands for fusion strategy, IAAF stands for intersection-aware adaptive fusion module, and ResCBAM stands for residual convolutional attention block. The feature resolution of the first ResCBAM block (residual convolutional attention block) is reduced by adjusting the stride from 1 to 2. This allows subsequent computations to operate at half the image size and combines the residuals of the pre-processing layer with downsampled gamma correction. This enhances the learnability of gamma correction and the local branch fusion strategy.

[0107] In addition, considering the standard gamma correction operation, such as the expression of point-by-point exponential operation. In the attention mechanism, by incorporating it directly into the attention process, this operation is adapted to the function within the CP module, so that it can be used as a global feature to estimate the optimal gamma correction for each feature channel. The modified attention output can be expressed as:

[0108] ;

[0109] Where, Represents gamma-corrected features and original features The attention features established, Indicates that based on the gamma correction feature, Represents the original feature of attention.

[0110] Furthermore, another residual is added to this attention mechanism to distinguish it from the reconstruction applied in the expression of the intersection-aware adaptive fusion module, making it an auxiliary attention supporting the residual features.

[0111] Adaptive gamma values, as an environmental factor representing the overall lighting conditions, may vary in different scenes. Therefore, this attention mechanism ensures that the network focuses on the areas where gamma correction produces the most significant enhancement in a wider range of images. This is the design principle of the CPGA block.

[0112] S2. Based on the pre-selected loss function, training dataset and evaluation indicators, the low-light image enhancement neural network is trained and evaluated.

[0113] In one embodiment, the loss function includes L1 loss, perceptual loss, HDR loss and SSIM loss;

[0114] When the low-light image enhancement neural network is trained using matching data, the training data set includes a League of Legends v1 (LOLv1) data set and a League of Legends v2 (LOLv2) data set, the LOLv2 data set includes a real capture subset and a synthetic subset, and the evaluation metrics include peak signal-to-noise ratio, structural similarity, and learnable perceptual image block similarity;

[0115] When the low-light image enhancement neural network is trained using unmatched data, the training data set includes a low-light image data set enhanced by illumination map estimation, a data set of a power-limited contrast enhancement technology for emissive displays based on histogram equalization, a naturalness enhancement algorithm data set for non-uniform illumination images, a cracked image enhancement and tone mapping algorithm data set, and a contrast enhancement technology data set based on hierarchical difference representation, and the evaluation index includes a natural image quality evaluator.

[0116] It should be explained that the present invention uses four loss functions, including L1 loss, perceptual loss, HDR loss and SSIM loss.

[0117] The L1 loss function is a commonly used loss function with good performance in image enhancement and restoration, and is defined as:

[0118] ;

[0119] Where, Indicates output, Indicates actual value.

[0120] Perceptual loss is a loss function commonly used in image restoration, style transfer, and image generation. It emphasizes capturing high-level features and structures that are very similar to human perception. The loss can be expressed as:

[0121] ;

[0122] Where, Represents the feature extractor of VGG16.

[0123] HDR L1 loss is calculated in the tone mapping domain, because HDR (high dynamic range) images are usually viewed after tone mapping. To achieve this, the µlaw function is generally used to calculate the loss:

[0124] ;

[0125] Where, represents the symbolic function, represents the logarithmic function, Indicates the use for normalizing input values. ux express and The product of u represents the µlaw parameter, T is the tone-mapped HDR image, x is the input image.

[0126] in, Used to perform nonlinear adjustments in the tone mapping process of an image. u is a constant that controls the nonlinear compression, µ Set it to 5000, and then apply the µlaw function to the L1 loss:

[0127] ;

[0128] Where, Representing images in the tone-mapped domain ensures that the loss is computed in a perceptually relevant space.

[0129] SSIM loss is a function that measures the similarity between two images based on structural information using the SSIM index (structural similarity index). It compares brightness, contrast, and structure, and better reflects perceptual quality than traditional pixel-wise loss. It can be written as:

[0130] ;

[0131] In addition, for paired data, this paper uses the League of Legends v1 (LOLv1) and League of Legends v2 (LOLv2) datasets. LOLv1 includes 485 images for training and testing, while LOLv2 includes two subsets: real capture and synthetic. The real capture subset (LOLv2 Real) has 689 images for training and 100 images for testing, while the synthetic subset (LOLv2 Synthetic) has 900 training images and 100 test images. PSNR, SSIM, and LPIPS are used as evaluation metrics.

[0132] For unpaired data, this paper uses five datasets: LIME, MEF, NPE, VV and DICM, and adopts NIQE as the evaluation indicator.

[0133] Among them, LOL (Deep Retinex Decomposition for Low-Light Enhancement): contains paired datasets of low-light / normal-light images;

[0134] PSNR (Peak Signal-to-Noise Ratio): Peak signal-to-noise ratio;

[0135] SSIM (Structural Similarity Index): structural similarity;

[0136] LPIPS (Learned Perceptual Image Patch Similarity): can learn to perceive image patch similarity;

[0137] LIME (Low-light Image Enhancement via Illumination Map Estimation): A dataset for enhancing low-light images through illumination map estimation.

[0138] MEF (Power-constrained contrast enhancement for emissive displays based on histogram equalization): a dataset of power-constrained contrast enhancement technology for emissive displays based on histogram equalization;

[0139] NPE (Naturalness preserved enhancement algorithm for non-uniform illumination images): a dataset of naturalness preserved enhancement algorithms for non-uniform illumination images;

[0140] VV (Busting image enhancement and tone-mapping algorithms: A collection of the most challenging cases from Vassilios Vonikakis): Busting image enhancement and tone-mapping algorithms: A collection of the most challenging cases from Vassilios Vonikakis;

[0141] DICM (Contrast enhancement based on layered difference representation): a dataset of contrast enhancement techniques based on layered difference representation;

[0142] NIQE (Natural Image Quality Evaluator): Natural image quality evaluator.

[0143] This paper will use several state-of-the-art methods such as LIME, Retinex-Net, KinD, EnGAN and the low-light image enhancement neural network CPGA-Net+ proposed in this paper to conduct experiments on benchmark datasets and evaluate the experimental results.

[0144] Table 1 Comparison of the present invention and SOTA methods on paired datasets

[0145]

[0146] Table 2 Comparison of image quality of the present invention on unpaired data

[0147]

[0148] According to the results in Tables 1 and 2, where the first and second place winners are indicated in bold and underlined, respectively, our method achieves a high standard compared to other methods, ranking third on paired datasets and first on five unpaired datasets, while maintaining a lightweight design with a low number of parameters and FLOPs. Furthermore, the method using the CPGA architecture achieves better quality on unpaired data, suggesting that the improved theoretical equation assumptions are closer to nature, resulting in more realistic images. Furthermore, while maintaining a lightweight design, our method successfully improves performance by 5% SSIM over CPGA-Net using the same architecture.

[0149] S3. Based on the trained low-light image enhancement neural network, ablation experiments are performed on the extended gamma correction attention module, the number of channel prior modules, and the loss function to obtain the enhanced image.

[0150] In one embodiment, the ablation experiment of the extended gamma correction attention module, the number of channel prior modules, and the loss function based on the trained low-light image enhancement neural network includes the following steps:

[0151] S31. Based on gamma correction and attention mechanism, an ablation experiment is conducted on the extended gamma correction attention module.

[0152] In one embodiment, performing an ablation experiment on the extended gamma correction attention module based on the gamma correction and the attention mechanism includes the following steps:

[0153] S311. Establish an attention mechanism based on gamma correction;

[0154] S312. Integrate gamma correction from the global branch to the local branch and perform ablation experiments on the extended gamma correction attention module.

[0155] S32. Based on the low-light image enhancement neural network, an ablation experiment was performed by changing the number of channel prior modules.

[0156] In one embodiment, performing an ablation experiment based on a low-light image enhancement neural network by changing the number of channel prior modules includes the following steps:

[0157] S321, based on the low-light image enhancement neural network, gradually changing the number of channel prior modules;

[0158] S322. Evaluate the model performance under each configuration separately and record the number of parameters and computational cost.

[0159] S33. Based on the default settings of the low-light image enhancement neural network, an ablation experiment is performed on the loss function.

[0160] In one embodiment, performing an ablation experiment on the loss function based on the default settings of the low-light image enhancement neural network includes the following steps:

[0161] S331. Based on the default settings of the low-light image enhancement neural network, ablation experiments are performed on the loss function respectively;

[0162] S332. Combined with the loss function after the ablation experiment, the best overall monitoring performance is obtained.

[0163] It should be explained that the present invention will analyze the effectiveness of each system module and training technique, including the design of CPGA blocks, the number of CP blocks and the loss function.

[0164] Table 3 Ablation study on CPGA module design

[0165]

[0166] Table 4 Ablation study on the number of CP modules

[0167]

[0168] Table 5 Ablation study of loss function

[0169]

[0170] First, as shown in Table 3, ablation experiments were conducted on the design of the CPGA block. The results show that gamma correction can be effectively integrated from the global branch to the local branch, improving overall performance and demonstrating the advantages of our method. By basing the attention mechanism on gamma correction, the enhancement process is consistent with the nonlinearities inherent in the imaging process and human perception. As shown in Table 4, ablation experiments were conducted on the number of CP blocks. The results show that increasing the number of CP blocks leads to improvement from 0 to 2 blocks, but there is no significant change from 2 to 4 blocks. However, with an increase in CP blocks, the number of parameters and computational cost (FLOPs) increase, resulting in greater computational requirements. Therefore, the optimal number of CP blocks should balance performance gains and resource efficiency. For the final design of our method, two CP blocks were selected to achieve a lightweight and efficient design.

[0171] Finally, as shown in Table 5, ablation experiments were conducted on the loss functions. L1 loss, perceptual loss, HDR L1 loss, and SSIM loss were tested. Results show that, using the default settings in CPGA-Net, the combination of L1 and perceptual losses performs well, improving PSNR by +2.82dB and SSIM by +2.4%. HDR L1 loss significantly enhances all three metrics, with a PSNR increase of +2.86dB, an SSIM increase of +4.1%, and a LPIPS decrease of 0.09. While SSIM loss achieves a 3.1% SSIM gain, it is less effective in improving PSNR. Ultimately, combining all of these losses leads to the best overall monitoring performance, achieving a PSNR of 4.04dB, an SSIM of 8.3%, and a LPIPS of -0.12.

[0172] According to another embodiment of the present invention, an image enhancement system based on a gamma correction prior-driven attention mechanism is also provided. Figure 2 As shown, the image enhancement system includes a neural network construction module 1, a neural network training module 2 and an ablation experiment module 3;

[0173] Neural network building module 1, used to build a low-light image enhancement neural network based on a pre-configured network structure;

[0174] Neural network training module 2, used to train and evaluate the low-light image enhancement neural network based on a pre-selected loss function, training data set and evaluation indicators;

[0175] Ablation experiment module 3 is used to perform ablation experiments on the extended gamma correction attention module, the number of channel prior modules and the loss function based on the trained low-light image enhancement neural network to obtain the enhanced image.

[0176] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An image enhancement method based on a gamma correction prior-driven attention mechanism, characterized in that: The image enhancement method comprises the following steps: S1. Build a low-light image enhancement neural network based on a pre-configured network structure. The low-light image enhancement neural network is constructed based on a preconfigured network structure, wherein the low-light image enhancement neural network includes a channel prior module and an extended gamma correction attention module, and includes the following steps: S11, channel prior module obtained based on Retinex theory and pre-configured atmospheric scattering model; S12. Introduce gamma correction in the attention process, and combine it with the channel prior module to use global features to estimate the optimal gamma correction of the feature channel to obtain the extended gamma correction attention module; S2. Train and evaluate the low-light image enhancement neural network based on the pre-selected loss function, training dataset, and evaluation metrics; S3. Based on the trained low-light image enhancement neural network, ablation experiments are conducted on the extended gamma correction attention module, the number of channel prior modules, and the loss function to obtain the enhanced image; The ablation experiment of the extended gamma correction attention module, the number of channel prior modules and the loss function based on the trained low-light image enhancement neural network includes the following steps: S31. Based on gamma correction and attention mechanism, an ablation experiment is conducted on the extended gamma correction attention module. The ablation experiment on the extended gamma correction attention module based on the gamma correction combined with the attention mechanism includes the following steps: S311. Establish an attention mechanism based on gamma correction; S312, integrate gamma correction from the global branch to the local branch, and perform ablation experiments on the extended gamma correction attention module; S32, based on the low-light image enhancement neural network, an ablation experiment was conducted by changing the number of channel prior modules; S33, based on the default settings of the low-light image enhancement neural network, conduct an ablation experiment on the loss function; The low-light image enhancement neural network includes a channel prior module for fusing original RGB channel features and an extended gamma correction attention module for estimating global features.

2. The image enhancement method based on gamma correction prior driven attention mechanism according to claim 1, characterized in that The channel prior module obtained based on Retinex theory and a pre-configured atmospheric scattering model includes the following steps: S111, couple the Retinex theory with the pre-configured atmospheric scattering model and transform it into a parameter learning model; S112. Use the convolutional network layer and the residual network layer to reconstruct the network structure expression of the parameter learning model to obtain the channel prior module.

3. The image enhancement method based on gamma correction prior driven attention mechanism according to claim 1, characterized in that The expressions of the channel prior module and the extended gamma correction attention module are: ； ； Where, represents the mapped input tensor with matching channels, represents the complement of atmospheric light intensity, represents atmospheric transmission, represents the input feature map, represents the channel prior characteristics, Represents gamma-corrected features and original features The attention features established, Indicates that based on the gamma correction feature, Represents the original feature of attention.

4. The image enhancement method based on gamma correction prior driven attention mechanism according to claim 1, characterized in that The loss functions include L1 loss, perceptual loss, HDR loss and SSIM loss; When the low-light image enhancement neural network is trained using matching data, the training data set includes a League of Legends v1 (LOLv1) data set and a League of Legends v2 (LOLv2) data set, the LOLv2 data set includes a real capture subset and a synthetic subset, and the evaluation metrics include peak signal-to-noise ratio, structural similarity, and learnable perceptual image block similarity; When the low-light image enhancement neural network is trained using unmatched data, the training data set includes a low-light image data set enhanced by illumination map estimation, a data set of a power-limited contrast enhancement technology for emissive displays based on histogram equalization, a naturalness enhancement algorithm data set for non-uniform illumination images, a cracked image enhancement and tone mapping algorithm data set, and a contrast enhancement technology data set based on hierarchical difference representation, and the evaluation index includes a natural image quality evaluator.

5. The image enhancement method based on gamma correction prior driven attention mechanism according to claim 1, characterized in that The ablation experiment based on the low-light image enhancement neural network by changing the number of channel prior modules includes the following steps: S321, based on the low-light image enhancement neural network, gradually changing the number of channel prior modules; S322. Evaluate the model performance under each configuration separately and record the number of parameters and computational cost.

6. The image enhancement method based on gamma correction prior driven attention mechanism according to claim 1, characterized in that The ablation experiment on the loss function based on the default settings of the low-light image enhancement neural network includes the following steps: S331. Based on the default settings of the low-light image enhancement neural network, ablation experiments are performed on the loss function respectively; S332. Combined with the loss function after the ablation experiment, the best overall monitoring performance is obtained.

7. An image enhancement system based on a gamma correction prior-driven attention mechanism, configured to implement the image enhancement method based on a gamma correction prior-driven attention mechanism according to any one of claims 1 to 6, characterized in that: The image enhancement system includes a neural network building module, a neural network training module and an ablation experiment module; A neural network building module for building a low-light image enhancement neural network based on a pre-configured network structure; A neural network training module is used to train and evaluate the low-light image enhancement neural network based on a pre-selected loss function, training dataset, and evaluation metrics; The ablation experiment module is used to perform ablation experiments on the extended gamma correction attention module, the number of channel prior modules, and the loss function based on the trained low-light image enhancement neural network to obtain the enhanced image.

Images